*What we do*\nExperify enables honest and authentic conversations around products

Experify

experify

Authors:
(1) Li Siyao, S-Lab, Nanyang Technological University;
(2) Tianpei Gu, Lexica and Work completed at UCLA;
(3) Weiye Xiao, Southeast University;
(4) Henghui Ding, S-Lab, Nanyang Technological University;
(5) Ziwei Liu, S-Lab, Nanyang Technological University;
(6) Chen Change Loy, S-Lab, Nanyang Technological University and a Corresponding Author. Table of Links Abstract and 1. Introduction Related Work
Mixamo Line Art Dataset
Our Approach
Experiment
Conclusion and References 5. Experiment Evaluation Metric. Following [16, 5], we adopt the chamfer distance (CD) as the evaluation metric, which has been initially introduced to measure the similarity between two point clouds. Formally, CD is computed as: 5.1. Comparison to Existing Methods As shown in Table 2, our AnimeInbet favorably outperforms all compared methods on both the validation set and the test set of MixamoLine240. On the validation set, our approach achieves an average CD value of 11.53, representing a significant improvement over the best-performing compared method, FILM, with over 30% enhancement. Upon closer inspection, the advantage of AnimeInbet becomes more pronounced as the frame gap increases (0.98, 5.72 and 9.47 for gaps of 1, 5, and 9, respectively), indicating that our method is more robust in handling larger motions. On the test set, our method maintains its lead over the other compared methods, with improvements of 0.70 (20%), 5.25 (29%), and 10.30 (31%) from the best-performing compared method FILM for the frame gaps of 1, 5, and 9, respectively. Given that both the characters and actions in the test set are new, our method’s superiority on the test set provides more convincing evidence of its advantages over the existing frame interpolation methods. To illustrate the advantages of our method, we present several inbetweening results in Figure 8. We arranged these examples in increasing levels of difficulty from top to bottom. When the motion is simple, compared methods can interpolate a relatively complete shape of the main body of the drawing. However, they tend to produce strong blurring (RIFE) or disappearance (VFIformer, EISAI, and FILM) of noticeable moving compositions (indicated by red arrows). In contrast, our method maintains a concise line structure in these key areas. When the input frames involve the whole body’s movement within large magnitudes, the intermediate frames predicted by the compared methods become indistinguishable and patchy, rendering the results invalid for further use. However, our AnimeInbet method can still preserve the general shape in the correct positions, even with a partial loss of details, which can be easily fixed with minor manual effort. User Study. To further evaluate the visual performance of our methods, we conduct a user study among 36 participants. For each participant, we randomly show 60 pairs, each composed of a result of AnimeInbet and that of a compared method, and ask the participant to select the better. To allow participants to take temporal consistency into the decision, we display these results in GIF formats formed by triplets of input frames and the inbetweened one. The winning rates of our method are shown in Figure 9, where AnimeInbet wins over 92% versus the compared methods. Notably, for “gap = 5” and “gap = 9” slots, the winning rates of our methods are close to 100% with smaller deviations than “gap = 1”, suggesting the advantages of our method on cases within large motions. 5.2. Ablation Study Embedding Features. To investigate the effectiveness of the three types of embeddings mentioned in Section 4.1, we trained several variants by removing the corresponding modules. As shown in Table 3, for each variant, we list the matching accuracy for all vertices (“Acc.”), the accuracy for non-occluded vertices (“Valid Acc.”) and the final CD values of inbetweening on the validation set (gap = 5). If removing the positional embedding EP , the “Valid Acc.” and the CD value drop 15.83% and 0.74, respectively; while the lacking of topological embedding ET lowers “Valid Acc.” by 5.66% and worsens CD by 0.43, which reveals the importance of these two components. Repositioning Propagation and Visibility Mask. We demonstrate the contribution of repositioning propagation (prepos. prop.) and visibility mask (vis. mask) both quantitatively and qualitatively. As shown in Table 4, without repositioning propagation, the CD value will be sharply worsened by 12.50 (112%), while the lacking of visibility mask will also make a drop of 1.69 (15%). An example is shown in Figure 10, where “w/o. repos. prop.” appears within many messy lines due to undefined positions for those unmatched vertices, while “w/o. vis. mask” shows some redundant segments (red box) after repositioning; the complete AnimeInbet can resolve these issues and produce a clean yet complete result. Geometrizor. As shown in Table 2, the quantitative metrics of AnimeInbet-VS are generally worse by around 0.6 compared to AnimeInbet. This is because VirtualSketcher [15] does not vectorize the line arts as precisely as the ground truth labels (average vertex number 587 vs 1,351). As shown in Figure 10, the curves in “AnimeInbet-VS” become sharper and lose some details, which decreases the quality of the inbetweened frame. Using a more accurate geometrizer would lead to higher quality inbetweening results for raster image inputs. Data Influence. As mentioned in Section 3, we created a validation set composed of 20 sequences of unseen characters but seen actions, 20 of unseen actions but seen characters and 4 of unseen both to explore the influence on data. Our experiment finds that whether the characters or the actions are seen does not fundamentally influence the inbetweening quality, while the motion magnitude is the key factor. As shown in Table 5, the CD value of unseen characters is 14.70, which is over 47% worse than that of unseen both due to larger vertex shifts (44.59 vs 29.62), while the difference between the CD values of unseen actions and unseen both is around 10% under similar occlusion rates and shifts. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Li Siyao, S-Lab, Nanyang Technological University; (2) Tianpei Gu, Lexica and Work completed at UCLA; (3) Weiye Xiao, Southeast University; (4) Henghui Ding, S-Lab, Nanyang Technological University; (5) Ziwei Liu, S-Lab, Nanyang Technological University; (6) Chen Change Loy, S-Lab, Nanyang Technological University and a Corresponding Author. Authors: Authors: (1) Li Siyao, S-Lab, Nanyang Technological University; (2) Tianpei Gu, Lexica and Work completed at UCLA; (3) Weiye Xiao, Southeast University; (4) Henghui Ding, S-Lab, Nanyang Technological University; (5) Ziwei Liu, S-Lab, Nanyang Technological University; (6) Chen Change Loy, S-Lab, Nanyang Technological University and a Corresponding Author. Table of Links Abstract and 1. Introduction Abstract and 1. Introduction Related Work Mixamo Line Art Dataset Our Approach Experiment Conclusion and References Related Work Related Work Mixamo Line Art Dataset Mixamo Line Art Dataset Our Approach Our Approach Experiment Experiment Conclusion and References Conclusion and References 5. Experiment Evaluation Metric. Following [16, 5], we adopt the chamfer distance (CD) as the evaluation metric, which has been initially introduced to measure the similarity between two point clouds. Formally, CD is computed as: Evaluation Metric. 5.1. Comparison to Existing Methods As shown in Table 2, our AnimeInbet favorably outperforms all compared methods on both the validation set and the test set of MixamoLine240. On the validation set, our approach achieves an average CD value of 11.53, representing a significant improvement over the best-performing compared method, FILM, with over 30% enhancement. Upon closer inspection, the advantage of AnimeInbet becomes more pronounced as the frame gap increases (0.98, 5.72 and 9.47 for gaps of 1, 5, and 9, respectively), indicating that our method is more robust in handling larger motions. On the test set, our method maintains its lead over the other compared methods, with improvements of 0.70 (20%), 5.25 (29%), and 10.30 (31%) from the best-performing compared method FILM for the frame gaps of 1, 5, and 9, respectively. Given that both the characters and actions in the test set are new, our method’s superiority on the test set provides more convincing evidence of its advantages over the existing frame interpolation methods. AnimeInbet To illustrate the advantages of our method, we present several inbetweening results in Figure 8. We arranged these examples in increasing levels of difficulty from top to bottom. When the motion is simple, compared methods can interpolate a relatively complete shape of the main body of the drawing. However, they tend to produce strong blurring (RIFE) or disappearance (VFIformer, EISAI, and FILM) of noticeable moving compositions (indicated by red arrows). In contrast, our method maintains a concise line structure in these key areas. When the input frames involve the whole body’s movement within large magnitudes, the intermediate frames predicted by the compared methods become indistinguishable and patchy, rendering the results invalid for further use. However, our AnimeInbet method can still preserve the general shape in the correct positions, even with a partial loss of details, which can be easily fixed with minor manual effort. AnimeInbet User Study. To further evaluate the visual performance of our methods, we conduct a user study among 36 participants. For each participant, we randomly show 60 pairs, each composed of a result of AnimeInbet and that of a compared method, and ask the participant to select the better. To allow participants to take temporal consistency into the decision, we display these results in GIF formats formed by triplets of input frames and the inbetweened one. The winning rates of our method are shown in Figure 9, where AnimeInbet wins over 92% versus the compared methods. Notably, for “gap = 5” and “gap = 9” slots, the winning rates of our methods are close to 100% with smaller deviations than “gap = 1”, suggesting the advantages of our method on cases within large motions. User Study. 5.2. Ablation Study Embedding Features. To investigate the effectiveness of the three types of embeddings mentioned in Section 4.1, we trained several variants by removing the corresponding modules. As shown in Table 3, for each variant, we list the matching accuracy for all vertices (“Acc.”), the accuracy for non-occluded vertices (“Valid Acc.”) and the final CD values of inbetweening on the validation set (gap = 5). If removing the positional embedding EP , the “Valid Acc.” and the CD value drop 15.83% and 0.74, respectively; while the lacking of topological embedding ET lowers “Valid Acc.” by 5.66% and worsens CD by 0.43, which reveals the importance of these two components. Repositioning Propagation and Visibility Mask. We demonstrate the contribution of repositioning propagation (prepos. prop.) and visibility mask (vis. mask) both quantitatively and qualitatively. As shown in Table 4, without repositioning propagation, the CD value will be sharply worsened by 12.50 (112%), while the lacking of visibility mask will also make a drop of 1.69 (15%). An example is shown in Figure 10, where “w/o. repos. prop.” appears within many messy lines due to undefined positions for those unmatched vertices, while “w/o. vis. mask” shows some redundant segments (red box) after repositioning; the complete AnimeInbet can resolve these issues and produce a clean yet complete result. Repositioning Propagation and Visibility Mask. Geometrizor . As shown in Table 2, the quantitative metrics of AnimeInbet-VS are generally worse by around 0.6 compared to AnimeInbet . This is because VirtualSketcher [15] does not vectorize the line arts as precisely as the ground truth labels (average vertex number 587 vs 1,351). As shown in Figure 10, the curves in “ AnimeInbet-VS ” become sharper and lose some details, which decreases the quality of the inbetweened frame. Using a more accurate geometrizer would lead to higher quality inbetweening results for raster image inputs. Geometrizor AnimeInbet-VS AnimeInbet VirtualSketcher AnimeInbet-VS Data Influence. As mentioned in Section 3, we created a validation set composed of 20 sequences of unseen characters but seen actions, 20 of unseen actions but seen characters and 4 of unseen both to explore the influence on data. Our experiment finds that whether the characters or the actions are seen does not fundamentally influence the inbetweening quality, while the motion magnitude is the key factor. As shown in Table 5, the CD value of unseen characters is 14.70, which is over 47% worse than that of unseen both due to larger vertex shifts (44.59 vs 29.62), while the difference between the CD values of unseen actions and unseen both is around 10% under similar occlusion rates and shifts. Data Influence. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

Deep Geometrized Cartoon Line Inbetweening: Experiment

Towards the Automation of Book Typesetting: Experimentation and Discussion

Authors:
(1) Han Jiang, HKUST and Equal contribution (hjiangav@connect.ust.hk);
(2) Haosen Sun, HKUST and Equal contribution (hsunas@connect.ust.hk);
(3) Ruoxuan Li, HKUST and Equal contribution (rliba@connect.ust.hk);
(4) Chi-Keung Tang, HKUST (cktang@cs.ust.hk);
(5) Yu-Wing Tai, Dartmouth College, (yu-wing.tai@dartmouth.edu). Table of Links Abstract and 1. Introduction 2. Related Work 2.1. NeRF Editing and 2.2. Inpainting Techniques 2.3. Text-Guided Visual Content Generation 3. Method 3.1. Training View Pre-processing 3.2. Progressive Training 3.3. 4D Extension 4. Experiments and 4.1. Experimental Setups 4.2. Ablation and comparison 5. Conclusion and 6. References 4. Experiments For our experiments, we select from dynamic scenes in the Nvidia Dynamic Scenes Dataset [35]. Scenes in this dataset are captured using a sparse set of 12 stationary cameras located in two rows, producing images of resolution 1015×1920. The static scenes we use are taken from one frame from the dynamic scenes. For the backbone NeRF, we use static and dynamic versions of K-Planes [25] implemented in nerfstudio [30]. For each scene, we conduct inpainting by replacing a foreground object with another text-prompted object with a different geometry. We will demonstrate the effectiveness of our method by showing the qualitative intermediate and final results. In addition, we will explain different parts of our design by ablations and comparisons on our baseline. 4.1. Qualitative results 3D Examples. We show several 3D inpainting examples in figure 2. For each individual inpainting task, we show 2 renderings of the final NeRF from different views to demonstrate the multiview consistency. Additionally, we show the first seed image, another pre-processed image, as well as the RGB and depth map in the three stages: before training, after warmup training, and after convergence. These beforeand-after images demonstrate the efficacy of each stage in our method. As shown in Figure 2, a roughly consistent preprocessed image can optimize a coarse inpainted NeRF after warmup training, and the geometry (represented by depth map) converges during warmup training. Then, fine convergence across views is achieved after the final training stage. All 3D inpainting tasks are trained on a single Nvidia RTX 4090 GPU. Warmup training takes approximately 0.5–1 hour, and the main training stage with IDU takes approximately 1–2 hours. 4D Example. We show a 4D inpainting example in figure 3 to demonstrate that our method has the potential to generalize to dynamic NeRFs. In this example, we remove the foreground object in the video of the seed view using E2FGVI [11], a flow-based method with optimization by feature propagation and content hallucination. For transferring motion to the generated object, after key point tracking, we estimate a rigid transformation between the key points, and propagate the pixels along the transformation. This dynamic scene consists of 16 frames, in which the first frame includes the first seed image. As shown in the figures, we successfully obtained an overall convergence on the generated object with correct motion for all the illustrated frames. This paper is available on arxiv under CC 4.0 license. Authors: (1) Han Jiang, HKUST and Equal contribution (hjiangav@connect.ust.hk); (2) Haosen Sun, HKUST and Equal contribution (hsunas@connect.ust.hk); (3) Ruoxuan Li, HKUST and Equal contribution (rliba@connect.ust.hk); (4) Chi-Keung Tang, HKUST (cktang@cs.ust.hk); (5) Yu-Wing Tai, Dartmouth College, (yu-wing.tai@dartmouth.edu). Authors: Authors: (1) Han Jiang, HKUST and Equal contribution (hjiangav@connect.ust.hk); (2) Haosen Sun, HKUST and Equal contribution (hsunas@connect.ust.hk); (3) Ruoxuan Li, HKUST and Equal contribution (rliba@connect.ust.hk); (4) Chi-Keung Tang, HKUST (cktang@cs.ust.hk); (5) Yu-Wing Tai, Dartmouth College, (yu-wing.tai@dartmouth.edu). Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2. Related Work 2.1. NeRF Editing and 2.2. Inpainting Techniques 2.1. NeRF Editing and 2.2. Inpainting Techniques 2.3. Text-Guided Visual Content Generation 2.3. Text-Guided Visual Content Generation 3. Method 3. Method 3.1. Training View Pre-processing 3.1. Training View Pre-processing 3.2. Progressive Training 3.2. Progressive Training 3.3. 4D Extension 3.3. 4D Extension 4. Experiments and 4.1. Experimental Setups 4. Experiments and 4.1. Experimental Setups 4.2. Ablation and comparison 4.2. Ablation and comparison 5. Conclusion and 6. References 5. Conclusion and 6. References 4. Experiments For our experiments, we select from dynamic scenes in the Nvidia Dynamic Scenes Dataset [35]. Scenes in this dataset are captured using a sparse set of 12 stationary cameras located in two rows, producing images of resolution 1015×1920. The static scenes we use are taken from one frame from the dynamic scenes. For the backbone NeRF, we use static and dynamic versions of K-Planes [25] implemented in nerfstudio [30]. For each scene, we conduct inpainting by replacing a foreground object with another text-prompted object with a different geometry. We will demonstrate the effectiveness of our method by showing the qualitative intermediate and final results. In addition, we will explain different parts of our design by ablations and comparisons on our baseline. 4.1. Qualitative results 3D Examples. We show several 3D inpainting examples in figure 2. For each individual inpainting task, we show 2 renderings of the final NeRF from different views to demonstrate the multiview consistency. Additionally, we show the first seed image, another pre-processed image, as well as the RGB and depth map in the three stages: before training, after warmup training, and after convergence. These beforeand-after images demonstrate the efficacy of each stage in our method. As shown in Figure 2, a roughly consistent preprocessed image can optimize a coarse inpainted NeRF after warmup training, and the geometry (represented by depth map) converges during warmup training. Then, fine convergence across views is achieved after the final training stage. All 3D inpainting tasks are trained on a single Nvidia RTX 4090 GPU. Warmup training takes approximately 0.5–1 hour, and the main training stage with IDU takes approximately 1–2 hours. 3D Examples. 4D Example. We show a 4D inpainting example in figure 3 to demonstrate that our method has the potential to generalize to dynamic NeRFs. In this example, we remove the foreground object in the video of the seed view using E2FGVI [11], a flow-based method with optimization by feature propagation and content hallucination. For transferring motion to the generated object, after key point tracking, we estimate a rigid transformation between the key points, and propagate the pixels along the transformation. This dynamic scene consists of 16 frames, in which the first frame includes the first seed image. As shown in the figures, we successfully obtained an overall convergence on the generated object with correct motion for all the illustrated frames. 4D Example. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

NeRF Editing and Inpainting Techniques: Experiments and Qualitative results

Authors:
(1) Omri Avrahami, Google Research and The Hebrew University of Jerusalem;
(2) Amir Hertz, Google Research;
(3) Yael Vinker, Google Research and Tel Aviv University;
(4) Moab Arar, Google Research and Tel Aviv University;
(5) Shlomi Fruchter, Google Research;
(6) Ohad Fried, Reichman University;
(7) Daniel Cohen-Or, Google Research and Tel Aviv University;
(8) Dani Lischinski, Google Research and The Hebrew University of Jerusalem. Table of Links Abstract and Introduction
Related Work
Method
Experiments
Limitations and Conclusions
A. Additional Experiments
B. Implementation Details
C. Societal Impact
References A. Additional Experiments Below, we provide additional experiments that were omitted from the main paper. In Appendix A.1 we provide additional comparisons and results of our method, and demonstrate its non-deterministic nature in Appendix A.2. In Appendix A.3 we compare our method against two na¨ıve baselines. Appendix A.4 presents the results of our method using different feature extractors. Lastly, in Appendix A.6 we provide results that reduce the concerns of dataset memorization by our method. A.1. Additional Comparisons and Results In Figure 9 we provide a qualitative comparison on the automatically generated prompts, and in Figure 10 we provide an additional qualitative comparison. Concurrently to our work, the DALL·E 3 model [12] was commercially released as part of the paid ChatGPT Plus [53] subscription, enabling generating images in a conversational setting. We tried, using a conversation, to create a consistent character of a Plasticine cat, as demonstrated in Figure 11. As can be seen, the generated characters share only some of the characteristics (e.g., big eyes) but not all of them (e.g., colors, textures and shapes). In Figure 12 we provide a qualitative comparison of the ablated cases. In addition, as demonstrated in Figure 13, our approach is applicable to consistent generation of a wide range of subjects, without the requirement for them to necessarily depict human characters or creatures. Figure 14 shows additional results of our method, demonstrating a variety of character styles. Lastly, in Figure 15 we demonstrate the ability of creating a fully consistent “life story” of a character using our method. A.2. Non-determinism of Our Method In Figures 16 and 17 we demonstrate the non-deterministic nature of our method. Using the same text prompt, we run our method multiple times with different initial seeds, thereby generating a different set of images for the identity clustering stage (Section 3.1). Consequently, the most cohesive cluster ccohesive is different in each run, yielding different consistent identities. This behavior of our method is aligned with the one-to-many nature of our task — a single text prompt may correspond to many identities. A.3. Naıve Baselines As explained in Section 4.1, we compared our method against a version of TI [20] and LoRA DB [71] that were trained on a single image (with a single identity). Instead, we could generate a small set of five images for the given prompt (that are not guaranteed to be of the same identity), and use this small dataset for TI and LoRA DB baselines, referred to as TI multi and LoRA DB multi, respectively. As can be seen in Figures 18 and 19, these baselines fail to achieve satisfactory identity consistency. A.4. Additional Feature Extractors Instead of using DINOv2 [54] features for the identity clustering stage (Section 3.1), we also experimented with two alternative feature extractors: DINOv1 [14] and CLIP [61] image encoder. We quantitatively evaluate our method with each of these feature extractors in terms of identity consistency and prompt similarity, as explained in Section 4.1. As can be seen in Figure 20, DINOv1 produces higher identity consistency, while sacrificing prompt similarity, whereas CLIP achieves higher prompt similarity at the expense of identity consistency. Qualitatively, as demonstrated in Figure 21, we found the DINOv1 extractor to perform similarly to DINOv2, whereas CLIP produces results with a slightly lower identity consistency. A.5. Additional Clustering Visualization In Figure 22 we provide a visualization of the clustering algorithm described in Section 3.1. As can be seen, given the input text prompt “a purple astronaut, digital art, smooth, sharp focus, vector art”, in the first iteration (top three rows), our algorithm divides the generated image set into three clusters: (1) focusing on the astronaut’s head, (2) an astronaut with no face, and (3) a full body astronaut. In the second iteration (bottom three rows), all the clusters share the same identity, that was extracted in the first iteration, as described in Section 3.2, and our algorithm divides them into clusters by their pose. A.6. Dataset Non-Memorization Our method is able to produce consistent characters, which raises the question of whether these characters already exist in the training data of the generative model. We employed SDXL [57] as our text-to-image model, whose training dataset is, unfortunately, undisclosed in the paper [57]. Consequently, we relied on the most likely overlapping dataset, LAION-5B [73], which was also utilized by Stable Diffusion V2. To probe for dataset memorization, we found the top 5 nearest neighbors in the dataset in terms of CLIP [61] image similarity, for a few representative characters from our paper, using an open-source solution [68]. As demonstrated in Figure 23, our method does not simply memorize images from the LAION-5B dataset. A.7. Stable Diffusion 2 Results We experimented with a version of our method that uses the Stable Diffusion 2 [69] model. The implementation is the same as explained in Appendix B.1, with the following changes: (1) The set of custom text embeddings τ in the character representation Θ (as explained in Section 2 in the main paper ), contains only one text embedding. (2) We used a higher learning rate of 5e-4. The rest of the implementation details are the same. More specifically, we used Stable Diffusion v2.1 implementation from Diffusers [86] library. As can be seen in Figure 24, when using the Stable Diffusion 2 backbone, our method can extract a consistent character, however, as expected, the results are of a lower quality than when using the SDXL [57] backbone that we use in the rest of this paper. This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. Authors: (1) Omri Avrahami, Google Research and The Hebrew University of Jerusalem; (2) Amir Hertz, Google Research; (3) Yael Vinker, Google Research and Tel Aviv University; (4) Moab Arar, Google Research and Tel Aviv University; (5) Shlomi Fruchter, Google Research; (6) Ohad Fried, Reichman University; (7) Daniel Cohen-Or, Google Research and Tel Aviv University; (8) Dani Lischinski, Google Research and The Hebrew University of Jerusalem. Authors: Authors: (1) Omri Avrahami, Google Research and The Hebrew University of Jerusalem; (2) Amir Hertz, Google Research; (3) Yael Vinker, Google Research and Tel Aviv University; (4) Moab Arar, Google Research and Tel Aviv University; (5) Shlomi Fruchter, Google Research; (6) Ohad Fried, Reichman University; (7) Daniel Cohen-Or, Google Research and Tel Aviv University; (8) Dani Lischinski, Google Research and The Hebrew University of Jerusalem. Table of Links Abstract and Introduction Related Work Method Experiments Limitations and Conclusions A. Additional Experiments B. Implementation Details C. Societal Impact References Abstract and Introduction Abstract and Introduction Related Work Related Work Method Method Experiments Experiments Limitations and Conclusions Limitations and Conclusions A. Additional Experiments A. Additional Experiments B. Implementation Details B. Implementation Details C. Societal Impact C. Societal Impact References References A. Additional Experiments Below, we provide additional experiments that were omitted from the main paper. In Appendix A.1 we provide additional comparisons and results of our method, and demonstrate its non-deterministic nature in Appendix A.2. In Appendix A.3 we compare our method against two na¨ıve baselines. Appendix A.4 presents the results of our method using different feature extractors. Lastly, in Appendix A.6 we provide results that reduce the concerns of dataset memorization by our method. A.1. Additional Comparisons and Results In Figure 9 we provide a qualitative comparison on the automatically generated prompts, and in Figure 10 we provide an additional qualitative comparison. Concurrently to our work, the DALL·E 3 model [12] was commercially released as part of the paid ChatGPT Plus [53] subscription, enabling generating images in a conversational setting. We tried, using a conversation, to create a consistent character of a Plasticine cat, as demonstrated in Figure 11. As can be seen, the generated characters share only some of the characteristics (e.g., big eyes) but not all of them (e.g., colors, textures and shapes). In Figure 12 we provide a qualitative comparison of the ablated cases. In addition, as demonstrated in Figure 13, our approach is applicable to consistent generation of a wide range of subjects, without the requirement for them to necessarily depict human characters or creatures. Figure 14 shows additional results of our method, demonstrating a variety of character styles. Lastly, in Figure 15 we demonstrate the ability of creating a fully consistent “life story” of a character using our method. A.2. Non-determinism of Our Method In Figures 16 and 17 we demonstrate the non-deterministic nature of our method. Using the same text prompt, we run our method multiple times with different initial seeds, thereby generating a different set of images for the identity clustering stage (Section 3.1). Consequently, the most cohesive cluster c cohesive is different in each run, yielding different consistent identities. This behavior of our method is aligned with the one-to-many nature of our task — a single text prompt may correspond to many identities. cohesive A.3. Naıve Baselines As explained in Section 4.1, we compared our method against a version of TI [20] and LoRA DB [71] that were trained on a single image (with a single identity). Instead, we could generate a small set of five images for the given prompt (that are not guaranteed to be of the same identity), and use this small dataset for TI and LoRA DB baselines, referred to as TI multi and LoRA DB multi, respectively. As can be seen in Figures 18 and 19, these baselines fail to achieve satisfactory identity consistency. A.4. Additional Feature Extractors Instead of using DINOv2 [54] features for the identity clustering stage (Section 3.1), we also experimented with two alternative feature extractors: DINOv1 [14] and CLIP [61] image encoder. We quantitatively evaluate our method with each of these feature extractors in terms of identity consistency and prompt similarity, as explained in Section 4.1. As can be seen in Figure 20, DINOv1 produces higher identity consistency, while sacrificing prompt similarity, whereas CLIP achieves higher prompt similarity at the expense of identity consistency. Qualitatively, as demonstrated in Figure 21, we found the DINOv1 extractor to perform similarly to DINOv2, whereas CLIP produces results with a slightly lower identity consistency. A.5. Additional Clustering Visualization In Figure 22 we provide a visualization of the clustering algorithm described in Section 3.1. As can be seen, given the input text prompt “a purple astronaut, digital art, smooth, sharp focus, vector art”, in the first iteration (top three rows), our algorithm divides the generated image set into three clusters: (1) focusing on the astronaut’s head, (2) an astronaut with no face, and (3) a full body astronaut. In the second iteration (bottom three rows), all the clusters share the same identity, that was extracted in the first iteration, as described in Section 3.2, and our algorithm divides them into clusters by their pose. A.6. Dataset Non-Memorization Our method is able to produce consistent characters, which raises the question of whether these characters already exist in the training data of the generative model. We employed SDXL [57] as our text-to-image model, whose training dataset is, unfortunately, undisclosed in the paper [57]. Consequently, we relied on the most likely overlapping dataset, LAION-5B [73], which was also utilized by Stable Diffusion V2. To probe for dataset memorization, we found the top 5 nearest neighbors in the dataset in terms of CLIP [61] image similarity, for a few representative characters from our paper, using an open-source solution [68]. As demonstrated in Figure 23, our method does not simply memorize images from the LAION-5B dataset. A.7. Stable Diffusion 2 Results We experimented with a version of our method that uses the Stable Diffusion 2 [69] model. The implementation is the same as explained in Appendix B.1, with the following changes: (1) The set of custom text embeddings τ in the character representation Θ (as explained in Section 2 in the main paper ), contains only one text embedding. (2) We used a higher learning rate of 5e-4. The rest of the implementation details are the same. More specifically, we used Stable Diffusion v2.1 implementation from Diffusers [86] library. As can be seen in Figure 24, when using the Stable Diffusion 2 backbone, our method can extract a consistent character, however, as expected, the results are of a lower quality than when using the SDXL [57] backbone that we use in the rest of this paper. This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. available on arxiv

The Chosen One: Consistent Characters in Text-to-Image Diffusion Models: Additional Experiments

Authors:
(1) Omri Avrahami, Google Research and The Hebrew University of Jerusalem;
(2) Amir Hertz, Google Research;
(3) Yael Vinker, Google Research and Tel Aviv University;
(4) Moab Arar, Google Research and Tel Aviv University;
(5) Shlomi Fruchter, Google Research;
(6) Ohad Fried, Reichman University;
(7) Daniel Cohen-Or, Google Research and Tel Aviv University;
(8) Dani Lischinski, Google Research and The Hebrew University of Jerusalem. Table of Links Abstract and Introduction
Related Work
Method
Experiments
Limitations and Conclusions
A. Additional Experiments
B. Implementation Details
C. Societal Impact
References 4. Experiments In Section 4.1 we compare our method against several baselines, both qualitatively and quantitatively. Next, in Section 4.2 we describe the user study we conducted and present its results. The results of an ablation study are reported in Section 4.3. Finally, in Section 4.4 we demonstrate several applications of our method. 4.1. Qualitative and Quantitative Comparison We compared our method against the most related personalization techniques [20, 42, 71, 89, 93]. In each experiment, each of these techniques is used to extract a character from a single image, generated by SDXL [57] from an input prompt p. The same prompt p is also provided as input to our method. Textual Inversion (TI) [20] optimizes a textual token using several images of the same concept, and we converted it to support SDXL by learning two text tokens, one for each of its text encoders, as we did in our method. In addition, we used LoRA DreamBooth [71] (LoRA DB), which we found less prone to overfitting than standard DB. Furthermore, we compared against all available image encoder techniques that encode a single image into the textual space of the diffusion model for later generation in novel contexts: BLIP-Diffusion [42], ELITE [89], and IP-adapter [93]. For all the baselines, we used the same prompt p to generate a single image, and used it to extract the identity via optimization (TI and LoRA DB) or encoding (ELITE, BLIP-diffusion and IP-adapter). In Figure 5 we qualitatively compare our method against the above baselines. While TI [20], BLIP-diffusion [42] and IP-adapter [93] are able to follow the specified prompt, they fail to produce a consistent character. LoRA DB [71] succeeds in consistent generation, but it does not always respond to the prompt. Furthermore, the resulting character is generated in the same fixed pose. ELITE [90] struggles with prompt following and the generated characters tend to be deformed. In comparison, our method is able to follow the prompt and maintain consistency, while generating appealing characters in different poses and viewing angles. In order to automatically evaluate our method and the baselines quantitatively, we instructed ChatGPT [53] to generate prompts for characters of different types (e.g., animals, creatures, objects, etc.) in different styles (e.g., stickers, animations, photorealistic images, etc.). Each of these prompts was then used to extract a consistent character by our method and by each of the baselines. Next, we generated these characters in a predefined collection of novel contexts. For a visual comparison, please refer to the supplementary material. We employ two standard evaluation metrics: prompt similarity and identity consistency, which are commonly used in the personalization literature [6, 20, 70]. Prompt similarity measures the correspondence between the generated images and the input text prompt. We use the standard CLIP [61] similarity, i.e., the normalized cosine similarity between the CLIP image embedding of the generated images and the CLIP text embedding of the source prompts. For measuring identity consistency, we calculate the similarity between the CLIP image embeddings of generated images of the same concept across different contexts. As can be seen in Figure 6 (left), there is an inherent trade-off between prompt similarity and identity consistency: LoRA DB and ELITE exhibit high identity consistency, while sacrificing prompt similarity. TI and BLIPdiffusion achieve high prompt similarity but low identity consistency. Our method and IP-adapter both lie on the Pareto front. However, our method achieves better identity consistency than IP-adapter, which is significant from the user’s perspective, as supported by our user study. 4.2. User Study We conducted a user study to evaluate our method, using the Amazon Mechanical Turk (AMT) platform [2]. We used the same generated prompts and samples that were used in Section 4.1 and asked the evaluators to rate the prompt similarity and identity consistency of each result on a Likert scale of 1–5. For ranking the prompt similarity, the evaluators were presented with the target text prompt and the result of our method and the baselines on the same page, and were asked to rate each of the images. For identity consistency, for each of the generated concepts, we compared our method and the baselines by randomly choosing pairs of generated images with different target prompts, and the evaluators were asked to rate on a scale of 1–5 whether the images contain the same main character. Again, all the pairs of the same character for the different baselines were shown on the same page. As can be seen in Figure 6 (right), our method again exhibits a good balance between identity consistency and prompt similarity, with a wider gap separating it from the baselines. For more details and statistical significance analbaselines. For more details and statistical significance analysis, read the supplementary material. 4.3. Ablation Study We conducted an ablation study for the following cases: (1) Without clustering — we omit the clustering step described in Section 3.1, and instead simply generate 5 images according to the input prompt. (2) Without LoRA — we reduce the optimizable representation Θ in the identity extraction stage, as described in Section 3.2, to consist of only the newly-added text tokens without the additional LoRA weights. (3) With re-initialization — instead of using the latest representation Θ in each of the optimization iterations, as described in Section 3.3, we re-initialize it in each iteration. (4) Single iteration — rather than iterating until convergence (Section 3.3), we stop after a single iteration. As can be seen in Figure 6 (left), all of the above key components are crucial for achieving a consistent identity in the final result: (1) removing the clustering harms the identity extraction stage because the training set is too diverse, (2) reducing the representation causes underfitting, as the model does not have enough parameters to properly capture the identity, (3) re-initializing the representation in each iteration, or (4) performing a single iteration, does not allow the model to converge into a single identity. For a visual comparison of the ablation study, as well as comparison of alternative feature extractors (DINOv1 [14] and CLIP [61]), please refer to the supplementary material. 4.4. Applications As demonstrated in Figure 7, our method can be used for various down-stream tasks, such as (a) Illustrating a story by breaking it into a different scenes and using the same consistent character for all of them. (b) Local text-driven image editing by integrating Blended Latent Diffusion [5, 7] — a consistent character can be injected into a specified location of a provided background image, in a novel pose specified by a text prompt. (c) Generating a consistent character with an additional pose control using ControlNet [97]. For more details, please refer to the supplementary material. This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. Authors: (1) Omri Avrahami, Google Research and The Hebrew University of Jerusalem; (2) Amir Hertz, Google Research; (3) Yael Vinker, Google Research and Tel Aviv University; (4) Moab Arar, Google Research and Tel Aviv University; (5) Shlomi Fruchter, Google Research; (6) Ohad Fried, Reichman University; (7) Daniel Cohen-Or, Google Research and Tel Aviv University; (8) Dani Lischinski, Google Research and The Hebrew University of Jerusalem. Authors: Authors: (1) Omri Avrahami, Google Research and The Hebrew University of Jerusalem; (2) Amir Hertz, Google Research; (3) Yael Vinker, Google Research and Tel Aviv University; (4) Moab Arar, Google Research and Tel Aviv University; (5) Shlomi Fruchter, Google Research; (6) Ohad Fried, Reichman University; (7) Daniel Cohen-Or, Google Research and Tel Aviv University; (8) Dani Lischinski, Google Research and The Hebrew University of Jerusalem. Table of Links Abstract and Introduction Related Work Method Experiments Limitations and Conclusions A. Additional Experiments B. Implementation Details C. Societal Impact References Abstract and Introduction Abstract and Introduction Related Work Related Work Method Method Experiments Experiments Limitations and Conclusions Limitations and Conclusions A. Additional Experiments A. Additional Experiments B. Implementation Details B. Implementation Details C. Societal Impact C. Societal Impact References References 4. Experiments In Section 4.1 we compare our method against several baselines, both qualitatively and quantitatively. Next, in Section 4.2 we describe the user study we conducted and present its results. The results of an ablation study are reported in Section 4.3. Finally, in Section 4.4 we demonstrate several applications of our method. 4.1. Qualitative and Quantitative Comparison We compared our method against the most related personalization techniques [20, 42, 71, 89, 93]. In each experiment, each of these techniques is used to extract a character from a single image, generated by SDXL [57] from an input prompt p. The same prompt p is also provided as input to our method. Textual Inversion (TI) [20] optimizes a textual token using several images of the same concept, and we converted it to support SDXL by learning two text tokens, one for each of its text encoders, as we did in our method. In addition, we used LoRA DreamBooth [71] (LoRA DB), which we found less prone to overfitting than standard DB. Furthermore, we compared against all available image encoder techniques that encode a single image into the textual space of the diffusion model for later generation in novel contexts: BLIP-Diffusion [42], ELITE [89], and IP-adapter [93]. For all the baselines, we used the same prompt p to generate a single image, and used it to extract the identity via optimization (TI and LoRA DB) or encoding (ELITE, BLIP-diffusion and IP-adapter). In Figure 5 we qualitatively compare our method against the above baselines. While TI [20], BLIP-diffusion [42] and IP-adapter [93] are able to follow the specified prompt, they fail to produce a consistent character. LoRA DB [71] succeeds in consistent generation, but it does not always respond to the prompt. Furthermore, the resulting character is generated in the same fixed pose. ELITE [90] struggles with prompt following and the generated characters tend to be deformed. In comparison, our method is able to follow the prompt and maintain consistency, while generating appealing characters in different poses and viewing angles. In order to automatically evaluate our method and the baselines quantitatively, we instructed ChatGPT [53] to generate prompts for characters of different types (e.g., animals, creatures, objects, etc.) in different styles (e.g., stickers, animations, photorealistic images, etc.). Each of these prompts was then used to extract a consistent character by our method and by each of the baselines. Next, we generated these characters in a predefined collection of novel contexts. For a visual comparison, please refer to the supplementary material. We employ two standard evaluation metrics: prompt similarity and identity consistency, which are commonly used in the personalization literature [6, 20, 70]. Prompt similarity measures the correspondence between the generated images and the input text prompt. We use the standard CLIP [61] similarity, i.e., the normalized cosine similarity between the CLIP image embedding of the generated images and the CLIP text embedding of the source prompts. For measuring identity consistency, we calculate the similarity between the CLIP image embeddings of generated images of the same concept across different contexts. As can be seen in Figure 6 (left), there is an inherent trade-off between prompt similarity and identity consistency: LoRA DB and ELITE exhibit high identity consistency, while sacrificing prompt similarity. TI and BLIPdiffusion achieve high prompt similarity but low identity consistency. Our method and IP-adapter both lie on the Pareto front. However, our method achieves better identity consistency than IP-adapter, which is significant from the user’s perspective, as supported by our user study. 4.2. User Study We conducted a user study to evaluate our method, using the Amazon Mechanical Turk (AMT) platform [2]. We used the same generated prompts and samples that were used in Section 4.1 and asked the evaluators to rate the prompt similarity and identity consistency of each result on a Likert scale of 1–5. For ranking the prompt similarity, the evaluators were presented with the target text prompt and the result of our method and the baselines on the same page, and were asked to rate each of the images. For identity consistency, for each of the generated concepts, we compared our method and the baselines by randomly choosing pairs of generated images with different target prompts, and the evaluators were asked to rate on a scale of 1–5 whether the images contain the same main character. Again, all the pairs of the same character for the different baselines were shown on the same page. As can be seen in Figure 6 (right), our method again exhibits a good balance between identity consistency and prompt similarity, with a wider gap separating it from the baselines. For more details and statistical significance analbaselines. For more details and statistical significance analysis, read the supplementary material. 4.3. Ablation Study We conducted an ablation study for the following cases: (1) Without clustering — we omit the clustering step described in Section 3.1, and instead simply generate 5 images according to the input prompt. (2) Without LoRA — we reduce the optimizable representation Θ in the identity extraction stage, as described in Section 3.2, to consist of only the newly-added text tokens without the additional LoRA weights. (3) With re-initialization — instead of using the latest representation Θ in each of the optimization iterations, as described in Section 3.3, we re-initialize it in each iteration. (4) Single iteration — rather than iterating until convergence (Section 3.3), we stop after a single iteration. As can be seen in Figure 6 (left), all of the above key components are crucial for achieving a consistent identity in the final result: (1) removing the clustering harms the identity extraction stage because the training set is too diverse, (2) reducing the representation causes underfitting, as the model does not have enough parameters to properly capture the identity, (3) re-initializing the representation in each iteration, or (4) performing a single iteration, does not allow the model to converge into a single identity. For a visual comparison of the ablation study, as well as comparison of alternative feature extractors (DINOv1 [14] and CLIP [61]), please refer to the supplementary material. 4.4. Applications As demonstrated in Figure 7, our method can be used for various down-stream tasks, such as (a) Illustrating a story by breaking it into a different scenes and using the same consistent character for all of them. (b) Local text-driven image editing by integrating Blended Latent Diffusion [5, 7] — a consistent character can be injected into a specified location of a provided background image, in a novel pose specified by a text prompt. (c) Generating a consistent character with an additional pose control using ControlNet [97]. For more details, please refer to the supplementary material. This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. available on arxiv

The Chosen One: Consistent Characters in Text-to-Image Diffusion Models: Experiments

Authors:
(1) Luyuan Peng, Acoustic Research Laboratory, National University of Singapore;
(2) Hari Vishnu, Acoustic Research Laboratory, National University of Singapore;
(3) Mandar Chitre, Acoustic Research Laboratory, National University of Singapore;
(4) Yuen Min Too, Acoustic Research Laboratory, National University of Singapore;
(5) Bharath Kalyan, Acoustic Research Laboratory, National University of Singapore;
(6) Rajat Mishra, Acoustic Research Laboratory, National University of Singapore. Table of Links I Introduction II Method III Datasets IV Experiments, Acknowledgment, and References IV. EXPERIMENTS We test three model configurations on both the simulator and tank datasets. First, we implement a pose regression model based on GoogLeNet [7] as our baseline model. Secondly, we implement a model using ResNet-50 [9] to evaluate the improvement possible by use of a deeper network. ResNet allows us to have deeper networks with more free parameters, and tries to avoid the problem of overfitting by having residual connections between layers [9]. Thirdly, we implement a pose regression model using an LSTM as an intermediate layer. The results indicate that all three configurations can perform well in both simulated and tank datasets as shown in Table I. The errors are minimal and of comparable magnitude to the noise in the pose recorded by the camera sensors. Fig. 4 shows that the trajectories predicted by the models are very close to the actual trajectories in terms of both position and orientation. With the simulator dataset, which is free of noise, turbidity, light distortion, and other challenges that may be observed in real underwater settings, the ResNet-50 and LSTM-based model performed better than the baseline model. On the other hand, with the tank datasets with some distortions typical of underwater scenarios, ResNet-50 performed worse than the baseline model, whereas the LSTM model did better than baseline with dataset #1 (which mostly featured translation but almost no rotation). This suggests the ResNet-50 architecture may be slightly overfitting despite the regularization, so it is not able to provide better performance in the tank dataset. We also note that data augmentation significantly improves the model performance, so in cases where there is no significant rotation, the use of data from both cameras can be used to bolster performance. Overall, these methods are robust for application in real underwater environments, and show promise for use in open water settings, where they will be tested next. ACKNOWLEDGMENT This research project is supported by A*STAR under its RIE2020 Advanced Manufacturing and Engineering (AME) Industry Alignment Fund - Pre-Positioning (IAF-PP) Grant No. A20H8a0241. REFERENCES [1] J. Gonzalez-Garcıa, A. Gomez-Espinosa, E. Cuan-Urquizo, L. G. ´ Garcıa-Valdovinos, T. Salgado-Jimenez, and J. A. E. Cabello, “Au-tonomous Underwater Vehicles: Localization, Navigation, and Commu- nication for Collaborative Missions,” Applied Sciences, vol. 10, no. 4, p. 1256, Feb. 2020. [2] L. Peng and M. Chitre, ”Regressing Poses from Monocular Images in an Underwater Environment”, 2022 OCEANS - Chennai, in press. [3] A. Kendall, M. Grimes and R. Cipolla, ”PoseNet: A Convolutional Network for Real-Time 6-DOF Camera Relocalization,” 2015 IEEE International Conference on Computer Vision (ICCV), 2015, pp. 2938- 2946, doi: 10.1109/ICCV.2015.336. [4] S. Hochreiter and J. Schmidhuber, “Long Short-Term Memory,” Neural Computation, vol. 9, no. 8, pp. 1735–1780, Nov. 1997, doi: 10.1162/neco.1997.9.8.1735. [5] F. Walch, C. Hazirbas, L. Leal-Taixe, T. Sattler, S. Hilsenbeck, and D. Cremers, “Image-based localization using lstms for structured feature correlation,” 2017 IEEE International Conference on Computer Vision (ICCV), 2017. [6] A. Kendall and R. Cipolla, ”Geometric Loss Functions for Camera Pose Regression with Deep Learning,” 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017, pp. 6555-6564, doi: 10.1109/CVPR.2017.694. [7] C. Szegedy et al., ”Going deeper with convolutions,” 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015, pp. 1-9, doi: 10.1109/CVPR.2015.7298594. [8] A. Chaudhary, R. Mishra, B. Kalyan, and M. Chitre, ”Development of an Underwater Simulator using Unity3D and Robot Operating Systems”, 2021 OCEANS - San Diego, in press. [9] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Luyuan Peng, Acoustic Research Laboratory, National University of Singapore; (2) Hari Vishnu, Acoustic Research Laboratory, National University of Singapore; (3) Mandar Chitre, Acoustic Research Laboratory, National University of Singapore; (4) Yuen Min Too, Acoustic Research Laboratory, National University of Singapore; (5) Bharath Kalyan, Acoustic Research Laboratory, National University of Singapore; (6) Rajat Mishra, Acoustic Research Laboratory, National University of Singapore. Authors: Authors: (1) Luyuan Peng, Acoustic Research Laboratory, National University of Singapore; (2) Hari Vishnu, Acoustic Research Laboratory, National University of Singapore; (3) Mandar Chitre, Acoustic Research Laboratory, National University of Singapore; (4) Yuen Min Too, Acoustic Research Laboratory, National University of Singapore; (5) Bharath Kalyan, Acoustic Research Laboratory, National University of Singapore; (6) Rajat Mishra, Acoustic Research Laboratory, National University of Singapore. Table of Links I Introduction I Introduction II Method II Method III Datasets III Datasets IV Experiments, Acknowledgment, and References IV Experiments, Acknowledgment, and References IV. EXPERIMENTS We test three model configurations on both the simulator and tank datasets. First, we implement a pose regression model based on GoogLeNet [7] as our baseline model. Secondly, we implement a model using ResNet-50 [9] to evaluate the improvement possible by use of a deeper network. ResNet allows us to have deeper networks with more free parameters, and tries to avoid the problem of overfitting by having residual connections between layers [9]. Thirdly, we implement a pose regression model using an LSTM as an intermediate layer. The results indicate that all three configurations can perform well in both simulated and tank datasets as shown in Table I. The errors are minimal and of comparable magnitude to the noise in the pose recorded by the camera sensors. Fig. 4 shows that the trajectories predicted by the models are very close to the actual trajectories in terms of both position and orientation. With the simulator dataset, which is free of noise, turbidity, light distortion, and other challenges that may be observed in real underwater settings, the ResNet-50 and LSTM-based model performed better than the baseline model. On the other hand, with the tank datasets with some distortions typical of underwater scenarios, ResNet-50 performed worse than the baseline model, whereas the LSTM model did better than baseline with dataset #1 (which mostly featured translation but almost no rotation). This suggests the ResNet-50 architecture may be slightly overfitting despite the regularization, so it is not able to provide better performance in the tank dataset. We also note that data augmentation significantly improves the model performance, so in cases where there is no significant rotation, the use of data from both cameras can be used to bolster performance. Overall, these methods are robust for application in real underwater environments, and show promise for use in open water settings, where they will be tested next. ACKNOWLEDGMENT This research project is supported by A*STAR under its RIE2020 Advanced Manufacturing and Engineering (AME) Industry Alignment Fund - Pre-Positioning (IAF-PP) Grant No. A20H8a0241. REFERENCES [1] J. Gonzalez-Garcıa, A. Gomez-Espinosa, E. Cuan-Urquizo, L. G. ´ Garcıa-Valdovinos, T. Salgado-Jimenez, and J. A. E. Cabello, “Au-tonomous Underwater Vehicles: Localization, Navigation, and Commu- nication for Collaborative Missions,” Applied Sciences, vol. 10, no. 4, p. 1256, Feb. 2020. [2] L. Peng and M. Chitre, ”Regressing Poses from Monocular Images in an Underwater Environment”, 2022 OCEANS - Chennai, in press. [3] A. Kendall, M. Grimes and R. Cipolla, ”PoseNet: A Convolutional Network for Real-Time 6-DOF Camera Relocalization,” 2015 IEEE International Conference on Computer Vision (ICCV), 2015, pp. 2938- 2946, doi: 10.1109/ICCV.2015.336. [4] S. Hochreiter and J. Schmidhuber, “Long Short-Term Memory,” Neural Computation, vol. 9, no. 8, pp. 1735–1780, Nov. 1997, doi: 10.1162/neco.1997.9.8.1735. [5] F. Walch, C. Hazirbas, L. Leal-Taixe, T. Sattler, S. Hilsenbeck, and D. Cremers, “Image-based localization using lstms for structured feature correlation,” 2017 IEEE International Conference on Computer Vision (ICCV), 2017. [6] A. Kendall and R. Cipolla, ”Geometric Loss Functions for Camera Pose Regression with Deep Learning,” 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017, pp. 6555-6564, doi: 10.1109/CVPR.2017.694. [7] C. Szegedy et al., ”Going deeper with convolutions,” 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015, pp. 1-9, doi: 10.1109/CVPR.2015.7298594. [8] A. Chaudhary, R. Mishra, B. Kalyan, and M. Chitre, ”Development of an Underwater Simulator using Unity3D and Robot Operating Systems”, 2021 OCEANS - San Diego, in press. [9] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Underwater Visual Localization Using Machine Learning and LSTM: Experiments, and References

AutoStory: Generating Diverse Storytelling Images with Minimal Human Effort: Experiments

After months of hard work, feedback, and learning, I launched Curl2Url on Product Hunt. Three months ago, on March 19, 2024, I eagerly submitted Curl2Url to Product Hunt, only to find it missing from the spotlight. Perplexed, I delved deeper, stumbling upon a tab housing all products posted that day, there was Curl2Url, lost at the bottom of the page. Even so, I would like to thank all the launch supporters. As a novice to the Product Hunt realm, this marked my inaugural product launch. I diligently absorbed insights from guides like "The Product Hunt Launch Guide" and Marc Lou's "How to Launch a Startup" (Jan. 24th, 2024), the latter securing the accolade of Maker of the Year at the Golden Kitty Awards. However, my immersion in success narratives inadvertently overshadowed other realities. That’s how I fall into the survivor bias. Guided by the belief that Monday to Thursday are good days if you want to get more traction, I chose Tuesday for Curl2Url's debut. Conversely, Friday to Sunday are your day targets if you want to launch the product just to get the Product Hunt badge. But if you don’t get featured, you barely have traction. Yet, amidst my immersion in triumph tales, I overlooked reads such as How Product Hunt really works ( Dec. 15th, 2015), How do you get featured on Product Hunt? (Sep. 19th, 2019) alongside insightful Hacker News comments on this post. Digging deeper into the purported algorithmic nuances dictating featured eligibility, I analyzed featured and non-featured products from April 2024. Checking out outlier days featured chances, I didn’t see a variable related to being featured or not. From my point of view, I saw products that needed improvements being featured and good products not being featured. Consequently, while intrigued by the inner workings of the algorithm I don’t want to spend more time checking how it works when the conclusion I extract from it is that it needs to improve. Conclusions and Future Last week, I took a fresh holiday that helped me to think about the experience and my future steps. While the experience thus far has been enriching, my focus moving forward lies in achieving sustainability. Having left my conventional job in January 2023, I have been living from savings. My goal is to find a job that gives me time for my projects, where I can learn from subjects I would like to delve into, with difficult challenges that keep me motivated and give me a good income. In light of this, I am committed to diversifying my approach and not relying solely on the prospects of a single platform. While Product Hunt holds promise, it is imperative to acknowledge the inherent unpredictability of such endeavors. I will continue working with Curl2Url for at least until August. It’s useful for me; there are several features I would like to add, and I can always relaunch it on Product Hunt. After months of hard work, feedback, and learning, I launched Curl2Url on Product Hunt . Three months ago, on March 19, 2024, I eagerly submitted Curl2Url to Product Hunt, only to find it missing from the spotlight. Perplexed, I delved deeper, stumbling upon a tab housing all products posted that day, there was Curl2Url, lost at the bottom of the page. Curl2Url Product Hunt Even so, I would like to thank all the launch supporters. As a novice to the Product Hunt realm, this marked my inaugural product launch. I diligently absorbed insights from guides like "The Product Hunt Launch Guide" and Marc Lou's "How to Launch a Startup" (Jan. 24th, 2024), the latter securing the accolade of Maker of the Year at the Golden Kitty Awards . However, my immersion in success narratives inadvertently overshadowed other realities. That’s how I fall into the survivor bias . "The Product Hunt Launch Guide" "How to Launch a Startup" Maker of the Year Golden Kitty Awards survivor bias Guided by the belief that Monday to Thursday are good days if you want to get more traction, I chose Tuesday for Curl2Url's debut. Conversely, Friday to Sunday are your day targets if you want to launch the product just to get the Product Hunt badge. But if you don’t get featured, you barely have traction. Yet, amidst my immersion in triumph tales, I overlooked reads such as How Product Hunt really works ( Dec. 15th, 2015), How do you get featured on Product Hunt? (Sep. 19th, 2019) alongside insightful Hacker News comments on this post . How Product Hunt really works How do you get featured on Product Hunt? this post Digging deeper into the purported algorithmic nuances dictating featured eligibility, I analyzed featured and non-featured products from April 2024. Checking out outlier days featured chances, I didn’t see a variable related to being featured or not. From my point of view, I saw products that needed improvements being featured and good products not being featured. Consequently, while intrigued by the inner workings of the algorithm I don’t want to spend more time checking how it works when the conclusion I extract from it is that it needs to improve. Conclusions and Future Last week, I took a fresh holiday that helped me to think about the experience and my future steps. While the experience thus far has been enriching, my focus moving forward lies in achieving sustainability. Having left my conventional job in January 2023, I have been living from savings. My goal is to find a job that gives me time for my projects, where I can learn from subjects I would like to delve into, with difficult challenges that keep me motivated and give me a good income. In light of this, I am committed to diversifying my approach and not relying solely on the prospects of a single platform. While Product Hunt holds promise, it is imperative to acknowledge the inherent unpredictability of such endeavors. I will continue working with Curl2Url for at least until August. It’s useful for me; there are several features I would like to add, and I can always relaunch it on Product Hunt.

Product Hunt Survivor Bias: My Experience

Today's standards in app development require developers to provide users with efficient and optimized experiences, which is crucial. The Distance Matrix API is an excellent tool that has emerged to enhance app functionality. By leveraging this API, developers can revolutionize how their apps interact with users. It will also provide information on real-time distance and travel time. This article covers the capabilities of the distance matrix API. It will cover its applications, benefits, and implementation strategies to highlight how it can transform app experiences. Understanding the Distance Matrix API The distance matrix API is a solution that provides travel time and distance data for multiple destinations. It calculates the distance and travel time between origins and destinations using various modes of transportation, including driving, walking, bicycling, and public transit. The API can manage multiple requests simultaneously, making it highly efficient for apps that require real-time information. Critical features of distance matrix API include: Multiple Modes of Transportation: The API supports different modes of transportation, allowing developers to offer versatile travel options.
Real-Time Data: Provides up-to-date travel times and distances, accounting for traffic conditions and other real-time factors.
Scalability: Handles large volumes of requests, making it suitable for apps with a high user base.
Flexibility: Can be integrated with other APIs and services, enhancing app functionality. Applications of the Distance Matrix API Navigation and mapping apps Navigation and mapping apps are among the primary beneficiaries of the Distance Matrix API. By integrating this API, these apps can offer users precise travel times and distances for multiple routes and transportation modes. For example, apps like Google Maps use the distance matrix API to provide accurate ETA and route options, improving user satisfaction and reliability. Ride-sharing services Ride-sharing services like Uber rely heavily on real-time distance and travel time data to match drivers with passengers efficiently. The distance matrix API helps these services calculate the quickest routes, estimate fares accurately, and minimize waiting times. These features ensure a smooth user experience. Delivery and logistics Optimizing routes and reducing delivery times are critical for delivery and logistics companies. The distance matrix API enables these companies to plan efficient routes, considering traffic conditions and other variables. This optimization leads to faster deliveries, reduced operational costs, and improved customer satisfaction. Travel and tourism Travel and tourism apps can use the distance matrix API to offer personalized travel itineraries. By calculating the travel time between various attractions, hotels, and restaurants, these apps can help users plan their trips more effectively, maximizing their time and enjoyment. Real estate Real estate apps can integrate the distance matrix API to inform potential buyers about properties' proximity to essential amenities. These basic amenities can include schools, hospitals, and shopping centers. This feature can significantly influence a buyer's decision, offering a competitive edge to real estate platforms. Fitness and health Fitness apps can use the distance matrix API to create personalized running, walking, or cycling routes for users. By offering real-time distance and travel time data, these apps can help users accomplish their fitness goals more efficiently and safely. Benefits of the Distance Matrix API Enhanced user experience The distance matrix API enhances the user experience by providing accurate, real-time information. Users can make appropriate decisions based on reliable data, leading to higher satisfaction and engagement with the app. Increased efficiency The API's ability to handle multiple requests simultaneously ensures that apps can provide information quickly and efficiently. This scalability is crucial for apps with a large user base, preventing slowdowns and ensuring smooth performance. Cost savings The distance matrix API can lead to significant cost savings for businesses. By optimizing routes and reducing travel times, companies can lower fuel costs, reduce vehicle wear and tear, and improve operational efficiency. Competitive advantage Integrating the distance matrix API can give apps a competitive edge by offering features that enhance user convenience and satisfaction. In a crowded app market, such differentiation can be crucial for attracting and retaining users. Customization The API's flexibility allows developers to customize its implementation according to their app's needs. Whether adjusting for different modes of transportation or integrating with other services, it can be tailored to fit diverse requirements. Best Practices Optimize requests: Use batch requests to minimize the number of API calls, reducing latency and costs.
Cache responses: Cache frequent responses to reduce API calls and improve performance. Ensure your cache is refreshed periodically to maintain data accuracy.
Monitor usage: Monitor your API usage to avoid exceeding limits and incurring additional costs. Use analytics tools to track and optimize usage patterns.
Ensure security: Secure your API key by restricting its usage to authorized domains or IP addresses. Rotate your API keys periodically to prevent unauthorized access.
Test thoroughly: Test your implementation to ensure it handles various scenarios, such as different transportation modes, multiple destinations, and real-time traffic conditions. Transforming User Experiences with Distance Matrix API The Distance Matrix API can revolutionize app experiences by providing real-time distance and travel time data. The API's applications are vast and varied, ranging from navigation and ride-sharing to delivery and fitness apps. The Distance Matrix API enhances user experience, increases efficiency, and offers a competitive edge, making it a valuable addition to any app developer's toolkit. As technology evolves, the API will likely see further enhancements and applications, making it an even more indispensable resource for app development. Today's standards in app development require developers to provide users with efficient and optimized experiences, which is crucial. The Distance Matrix API is an excellent tool that has emerged to enhance app functionality. By leveraging this API, developers can revolutionize how their apps interact with users. It will also provide information on real-time distance and travel time. This article covers the capabilities of the distance matrix API. It will cover its applications, benefits, and implementation strategies to highlight how it can transform app experiences. Understanding the Distance Matrix API The distance matrix API is a solution that provides travel time and distance data for multiple destinations. It calculates the distance and travel time between origins and destinations using various modes of transportation, including driving, walking, bicycling, and public transit. The API can manage multiple requests simultaneously, making it highly efficient for apps that require real-time information. Critical features of distance matrix API include: Multiple Modes of Transportation: The API supports different modes of transportation, allowing developers to offer versatile travel options. Real-Time Data: Provides up-to-date travel times and distances, accounting for traffic conditions and other real-time factors. Scalability: Handles large volumes of requests, making it suitable for apps with a high user base. Flexibility: Can be integrated with other APIs and services, enhancing app functionality. Multiple Modes of Transportation: The API supports different modes of transportation, allowing developers to offer versatile travel options. Real-Time Data: Provides up-to-date travel times and distances, accounting for traffic conditions and other real-time factors. Scalability: Handles large volumes of requests, making it suitable for apps with a high user base. Flexibility: Can be integrated with other APIs and services, enhancing app functionality. Applications of the Distance Matrix API Navigation and mapping apps Navigation and mapping apps are among the primary beneficiaries of the Distance Matrix API. By integrating this API, these apps can offer users precise travel times and distances for multiple routes and transportation modes. Navigation and mapping apps Navigation and mapping apps For example, apps like Google Maps use the distance matrix API to provide accurate ETA and route options, improving user satisfaction and reliability. Ride-sharing services Ride-sharing services like Uber rely heavily on real-time distance and travel time data to match drivers with passengers efficiently. The distance matrix API helps these services calculate the quickest routes, estimate fares accurately, and minimize waiting times. These features ensure a smooth user experience. Delivery and logistics Optimizing routes and reducing delivery times are critical for delivery and logistics companies. The distance matrix API enables these companies to plan efficient routes, considering traffic conditions and other variables. This optimization leads to faster deliveries, reduced operational costs, and improved customer satisfaction. Travel and tourism Travel and tourism apps can use the distance matrix API to offer personalized travel itineraries. By calculating the travel time between various attractions, hotels, and restaurants, these apps can help users plan their trips more effectively, maximizing their time and enjoyment. Real estate Real estate apps can integrate the distance matrix API to inform potential buyers about properties' proximity to essential amenities. These basic amenities can include schools, hospitals, and shopping centers. Real estate apps Real estate apps This feature can significantly influence a buyer's decision, offering a competitive edge to real estate platforms. Fitness and health Fitness apps can use the distance matrix API to create personalized running, walking, or cycling routes for users. By offering real-time distance and travel time data, these apps can help users accomplish their fitness goals more efficiently and safely. Benefits of the Distance Matrix API Enhanced user experience The distance matrix API enhances the user experience by providing accurate, real-time information. Users can make appropriate decisions based on reliable data, leading to higher satisfaction and engagement with the app. distance matrix API Increased efficiency The API's ability to handle multiple requests simultaneously ensures that apps can provide information quickly and efficiently. This scalability is crucial for apps with a large user base, preventing slowdowns and ensuring smooth performance. Cost savings The distance matrix API can lead to significant cost savings for businesses. By optimizing routes and reducing travel times, companies can lower fuel costs, reduce vehicle wear and tear, and improve operational efficiency. Competitive advantage Integrating the distance matrix API can give apps a competitive edge by offering features that enhance user convenience and satisfaction. In a crowded app market, such differentiation can be crucial for attracting and retaining users. Customization The API's flexibility allows developers to customize its implementation according to their app's needs. Whether adjusting for different modes of transportation or integrating with other services, it can be tailored to fit diverse requirements. Best Practices Optimize requests: Use batch requests to minimize the number of API calls, reducing latency and costs. Cache responses: Cache frequent responses to reduce API calls and improve performance. Ensure your cache is refreshed periodically to maintain data accuracy. Monitor usage: Monitor your API usage to avoid exceeding limits and incurring additional costs. Use analytics tools to track and optimize usage patterns. Ensure security: Secure your API key by restricting its usage to authorized domains or IP addresses. Rotate your API keys periodically to prevent unauthorized access. Test thoroughly: Test your implementation to ensure it handles various scenarios, such as different transportation modes, multiple destinations, and real-time traffic conditions. Optimize requests: Use batch requests to minimize the number of API calls, reducing latency and costs. Cache responses: Cache frequent responses to reduce API calls and improve performance. Ensure your cache is refreshed periodically to maintain data accuracy. Monitor usage: Monitor your API usage to avoid exceeding limits and incurring additional costs. Use analytics tools to track and optimize usage patterns. Ensure security: Secure your API key by restricting its usage to authorized domains or IP addresses. Rotate your API keys periodically to prevent unauthorized access. Test thoroughly: Test your implementation to ensure it handles various scenarios, such as different transportation modes, multiple destinations, and real-time traffic conditions. Transforming User Experiences with Distance Matrix API The Distance Matrix API can revolutionize app experiences by providing real-time distance and travel time data. The API's applications are vast and varied, ranging from navigation and ride-sharing to delivery and fitness apps. The Distance Matrix API enhances user experience, increases efficiency, and offers a competitive edge, making it a valuable addition to any app developer's toolkit. As technology evolves, the API will likely see further enhancements and applications, making it an even more indispensable resource for app development.

Revolutionizing App Experiences With Distance Matrix API

Authors:
(1) Enis Chenchene, Department of Mathematics and Scientific Computing, University of Graz;
(2) Hui Huang, Department of Mathematics and Scientific Computing, University of Graz;
(3) Jinniao Qiu, Department of Mathematics and Statistics, University of Calgary. Table of Links Abstract and 1 Introduction 2 Global convergence 2.1 Quantitative Laplace principle 2.2 Global convergence in mean-field law 3 Numerical experiments and 3.1 One-dimensional illustrative example 3.2 Nonlinear oligopoly games with several goods 4 Conclusion, Acknowledgments, Appendix, and References 3 Numerical experiments In this section, we present our numerical experiments, which are performed in Python on a 12thGen. Intel(R) Core(TM) i7–1255U, 1.70–4.70 GHz laptop with 16 Gb of RAM and are available for reproducibility at https://github.com/echnen/CBO-multiplayer. As usual in CBO schemes, we discretize the interacting particle system in (1.2) with a Euler– Maruyama time discretization scheme [34], resulting in the method depicted in Algorithm 1. 3.1 One-dimensional illustrative example 3.1.1 Experimental setup 3.1.2 Results and discussion This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Enis Chenchene, Department of Mathematics and Scientific Computing, University of Graz; (2) Hui Huang, Department of Mathematics and Scientific Computing, University of Graz; (3) Jinniao Qiu, Department of Mathematics and Statistics, University of Calgary. Authors: Authors: (1) Enis Chenchene, Department of Mathematics and Scientific Computing, University of Graz; (2) Hui Huang, Department of Mathematics and Scientific Computing, University of Graz; (3) Jinniao Qiu, Department of Mathematics and Statistics, University of Calgary. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Global convergence 2 Global convergence 2.1 Quantitative Laplace principle 2.1 Quantitative Laplace principle 2.2 Global convergence in mean-field law 2.2 Global convergence in mean-field law 3 Numerical experiments and 3.1 One-dimensional illustrative example 3 Numerical experiments and 3.1 One-dimensional illustrative example 3.2 Nonlinear oligopoly games with several goods 3.2 Nonlinear oligopoly games with several goods 4 Conclusion, Acknowledgments, Appendix, and References 4 Conclusion, Acknowledgments, Appendix, and References 3 Numerical experiments In this section, we present our numerical experiments, which are performed in Python on a 12thGen. Intel(R) Core(TM) i7–1255U, 1.70–4.70 GHz laptop with 16 Gb of RAM and are available for reproducibility at https://github.com/echnen/CBO-multiplayer. As usual in CBO schemes, we discretize the interacting particle system in (1.2) with a Euler– Maruyama time discretization scheme [34], resulting in the method depicted in Algorithm 1. 3.1 One-dimensional illustrative example 3.1.1 Experimental setup 3.1.1 Experimental setup 3.1.2 Results and discussion 3.1.2 Results and discussion This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

A Consensus-Based Algorithm for Non-Convex Multiplayer Games: Numerical Experiments

SEC v. Consensys Software Inc. Court Filing, retrieved on June 28, 2024, is part of HackerNoon’s Legal PDF Series. You can jump to any part in this filing here. This part is 10 of 26. A. Consensys Solicits Investors And Holds Itself Out As A Place To Buy and Sell Crypto Asset Securities Through Its MetaMask Swaps Brokerage Service. 48. On October 6, 2020, Consensys released MetaMask Swaps to the public. 49. Since then, Consensys has advertised and promoted MetaMask Swaps on its MetaMask website, metamask.io, and social media such as X (formerly, Twitter) as a way for investors to participate in the markets for crypto assets, including crypto asset securities. 50. In its October 6, 2020, launch announcement, Consensys claimed that MetaMask Swaps “offers [investors] the best trading experience in Defi” (referring to so-called “decentralized finance”). 51. More specifically, Consensys stated that the MetaMask Swaps platform offered investors the “best prices” and “deepest liquidity.” 52. The announcement concluded: “MetaMask enables more trades more efficiently by providing an optimized path for every trade.” 53. Consensys conveyed similar promotional messages on its website, at least as of December 2023 54. Its website stated, “Find the best price every time. Swaps ensures that you always have access to the largest selection of tokens and the most competitive prices.” 55. Its website continued to explain that, as part of its efforts to “locate the best trade,” Consensys “sources the best prices and determines which liquidity source is the most gas efficient.” (“Gas” is the fee that investors pay to record a transaction on the Ethereum blockchain. Gas fees are demand-based and increase as more users seek to use the platform. The higher the gas price paid for the transaction, the faster the transaction is likely to be mined. Gas is not refundable even if a transaction fails.) 56. In a January 2022 blog post, Consensys touted MetaMask Swaps as “The Optimized DeFi Trading Experience” and argued that its methodology increased the likelihood of best execution. 57. Consensys also maintains a social media accounts on X called “@MetaMask” and “@MetaMaskSupport” through which it has promoted MetaMask Swaps, apprised investors of new platform developments, and provided customer support. 58. For example, on December 21, 2022, Consensys posted on X, “We’ve expanded the Swaps experience,” including a link to a MetaMask news release on its website, which touted the expansion of the MetaMask Swaps platform capabilities to additional networks. 59. In another December 21, 2022, post on X, Consensys targeted potential investors who may have been unfamiliar with MetaMask Swaps. It stated, “Wait – what even is a Swap?! We got you. Learn all about MetaMask Swaps here.” The accompanying image stated that MetaMask is “[t]he best way to swap your digital tokens” and “Effortless Crypto Swapping and Token Exchange on MetaMask.” 60. Consensys posted another “news” release promoting MetaMask Swaps on its website on July 17, 2023. In a section titled “Unrivaled convenience,” Consensys stated, “Swaps is undeniably convenient. You never have to leave MetaMask to execute your transaction through our vetted group of providers.” The release continued, “Swaps ensures you receive . . . the maximum value for your transaction.” 61. On its MetaMask website and Consensys’s YouTube channel, Consensys has also published a video about its MetaMask Swaps service in which it casts MetaMask Swaps as, in essence, a broker of crypto assets for investors: “Introducing MetaMask Swaps . . . The easiest way to trade Ethereum tokens right from inside your MetaMask Wallet.” This video has been publicly available since March 2021. 62. The video touts the user-friendly nature of the MetaMask Swaps platform and explains that “what makes the [MetaMask Swaps platform] so useful is what it’s doing behind the scenes.” 63. The video narration states (referring to what in the crypto asset markets are called “decentralized exchanges” or “DEXs”) that MetaMask Swaps “compares various DEXs, aggregators, and market makers to find you the best price, with the lowest network fees, and the least slippage.” 64. The video also notes that MetaMask Swaps can split the investor’s trade up “among several providers to give [the investor] access to greater combined liquidity.” 65. The video concludes that MetaMask Swaps does “all that, without you having to, well, think about it, really.” Continue Reading Here. About HackerNoon Legal PDF Series: We bring you the most important technical and insightful public domain court case filings.
This court case retrieved on June 28, 2024, storage.courtlistener.com is part of the public domain. The court-created documents are works of the federal government, and under copyright law, are automatically placed in the public domain and may be shared without legal restriction. SEC v. Consensys Software Inc. Court Filing, retrieved on June 28, 2024, is part of HackerNoon’s Legal PDF Series. You can jump to any part in this filing here. This part is 10 of 26. SEC v. Consensys Software Inc. Court Filing, retrieved on June 28, 2024, is part of HackerNoon’s Legal PDF Series . You can jump to any part in this filing here . This part is 10 of 26. HackerNoon’s Legal PDF Series here A. Consensys Solicits Investors And Holds Itself Out As A Place To Buy and Sell Crypto Asset Securities Through Its MetaMask Swaps Brokerage Service. 48. On October 6, 2020, Consensys released MetaMask Swaps to the public. 49. Since then, Consensys has advertised and promoted MetaMask Swaps on its MetaMask website, metamask.io, and social media such as X (formerly, Twitter) as a way for investors to participate in the markets for crypto assets, including crypto asset securities. 50. In its October 6, 2020, launch announcement, Consensys claimed that MetaMask Swaps “offers [investors] the best trading experience in Defi” (referring to so-called “decentralized finance”). 51. More specifically, Consensys stated that the MetaMask Swaps platform offered investors the “best prices” and “deepest liquidity.” 52. The announcement concluded: “MetaMask enables more trades more efficiently by providing an optimized path for every trade.” 53. Consensys conveyed similar promotional messages on its website, at least as of December 2023 54. Its website stated, “Find the best price every time. Swaps ensures that you always have access to the largest selection of tokens and the most competitive prices.” 55. Its website continued to explain that, as part of its efforts to “locate the best trade,” Consensys “sources the best prices and determines which liquidity source is the most gas efficient.” (“Gas” is the fee that investors pay to record a transaction on the Ethereum blockchain. Gas fees are demand-based and increase as more users seek to use the platform. The higher the gas price paid for the transaction, the faster the transaction is likely to be mined. Gas is not refundable even if a transaction fails.) 56. In a January 2022 blog post, Consensys touted MetaMask Swaps as “The Optimized DeFi Trading Experience” and argued that its methodology increased the likelihood of best execution. 57. Consensys also maintains a social media accounts on X called “@MetaMask” and “@MetaMaskSupport” through which it has promoted MetaMask Swaps, apprised investors of new platform developments, and provided customer support. 58. For example, on December 21, 2022, Consensys posted on X, “We’ve expanded the Swaps experience,” including a link to a MetaMask news release on its website, which touted the expansion of the MetaMask Swaps platform capabilities to additional networks. 59. In another December 21, 2022, post on X, Consensys targeted potential investors who may have been unfamiliar with MetaMask Swaps. It stated, “Wait – what even is a Swap?! We got you. Learn all about MetaMask Swaps here.” The accompanying image stated that MetaMask is “[t]he best way to swap your digital tokens” and “Effortless Crypto Swapping and Token Exchange on MetaMask.” 60. Consensys posted another “news” release promoting MetaMask Swaps on its website on July 17, 2023. In a section titled “Unrivaled convenience,” Consensys stated, “Swaps is undeniably convenient. You never have to leave MetaMask to execute your transaction through our vetted group of providers.” The release continued, “Swaps ensures you receive . . . the maximum value for your transaction.” 61. On its MetaMask website and Consensys’s YouTube channel, Consensys has also published a video about its MetaMask Swaps service in which it casts MetaMask Swaps as, in essence, a broker of crypto assets for investors: “Introducing MetaMask Swaps . . . The easiest way to trade Ethereum tokens right from inside your MetaMask Wallet.” This video has been publicly available since March 2021. 62. The video touts the user-friendly nature of the MetaMask Swaps platform and explains that “what makes the [MetaMask Swaps platform] so useful is what it’s doing behind the scenes.” 63. The video narration states (referring to what in the crypto asset markets are called “decentralized exchanges” or “DEXs”) that MetaMask Swaps “compares various DEXs, aggregators, and market makers to find you the best price, with the lowest network fees, and the least slippage.” 64. The video also notes that MetaMask Swaps can split the investor’s trade up “among several providers to give [the investor] access to greater combined liquidity.” 65. The video concludes that MetaMask Swaps does “all that, without you having to, well, think about it, really.” Continue Reading Here. Continue Reading Here . Here About HackerNoon Legal PDF Series: We bring you the most important technical and insightful public domain court case filings. This court case retrieved on June 28, 2024, storage.courtlistener.com is part of the public domain. The court-created documents are works of the federal government, and under copyright law, are automatically placed in the public domain and may be shared without legal restriction. About HackerNoon Legal PDF Series: We bring you the most important technical and insightful public domain court case filings. This court case retrieved on June 28, 2024, storage.courtlistener.com is part of the public domain. The court-created documents are works of the federal government, and under copyright law, are automatically placed in the public domain and may be shared without legal restriction. storage.courtlistener.com

MetaMask Swaps Marketed as 'Best' Crypto Trading Experience, Faces SEC Charges

React Native has been a hot topic among developers since its launch by Meta in 2015. Want to know the reason? It’s pretty simple—“Ultimate UX.” Yes, it offers UX that you, as a developer or entrepreneur, have never experienced. But your users can! HOW? I’ll tell you all about it and something more about React Native, which can help you achieve the ultimate UX with low cost and effort. Let's begin by understanding why it’s such a magnet attracting app developers worldwide. This framework allowed developers to build apps using a single code across platforms. This saved time and money by reducing the need to create separate codes for each platform. You might be skeptical about the quality of the apps in such cases. What if I tell you there is a way to ensure quality, optimize cost, and provide enhanced UX? Does that seem far from the truth? Well, not really! I recently came across React Native Libraries, which are easy to set up, take less time, and save money. Want to know which one of them is best for you? Here is a list! Top 7 React Native Libraries For Ultimate UX! Here are the top Libraries that will help you build a React Native App with a great user experience: 1. React Native Elements React Native Elements is a UI component library for building apps with better navigation and screen space. It has over 24,700 stars on GitHub. Navigation components: Gets components like badges and avatars, headers and tabs, and features for handling user interactions like checkboxes and switches.
Screen space: This feature provides tools for creating dialogues and bottom sheets, allowing for additional content without taking up too much screen space.
Community and updates: Its active and regular community ensures the app stays updated with the latest design trends and best practices. 2. React Native Paper React Native Paper gets over 10,000 stars on GitHub. It focuses on providing a consistent look for your app across all platforms (iOS and Android). The best part, it follows Google's Material Design guidelines. Now, for those unfamiliar with material design let me explain in brief. Material design refers to print design methods that include typography, grids, space, scale, color, and imagery. It helps to create a hierarchy, meaning, and focus that gives users an immersive experience. Material design style: Employs Material design styles and themes, ensuring that apps developed using React Native Paper have a clean and professional appearance. Dark Mode: Supports dark mode, which can significantly enhance the user experience, especially in low-light environments. Customization: Customizable style of components to match the specific needs of the brand identity while maintaining consistency. API Integration: Technically, React Native Paper leverages native platform APIs for optimal performance and a natural feel. 3. React Native Vector Icons React Native Vector Icons specializes in icons and boasts over 3,000 appealing and scalable icons that can be used in your React Native app. It has over 15,000 starts on GitHub. Icons Collection: The library bundles a massive collection of icons from popular icon sets, such as FontAwesome, Entypo, ion icons, Material Icons, and more.
Crisp rendering: These scalable icons extend, style, and integrate seamlessly into your project, ensuring crisp rendering across all screen resolutions without quality loss.
Flexibility: These icons are versatile, functioning as emojis alongside text, button components, or native image elements.
SVG Support: It supports SVG via Fontello or regular icon fonts, making it easy to customize the appearance of the icons.
Time saver: The library is handy for saving time by not having to create custom icons from scratch. 4. React Native Navigation React Native Navigation is a library that allows the creation of complex navigation flows easily and provides features like: Stack navigation (used for hierarchical navigation),
Drawer navigation (for side-menu navigation),
Tab navigators (for bottom tab navigation) and
Modal navigators (for overlays). Other crucial features are, Animation and Transitions: Offers customizable navigation features like animations and transitions, and it uses an API to manage them and the user journey as well.
Supports Deep Linking: Allows users to navigate specific parts of your app directly through external links. 5. React Native Gesture Handler Users are increasingly looking for gesture-based interaction in mobile apps, and this library is specially designed for it. React Native Gesture Handler lets developers create responsive gesture-based interactions in their React Native apps. Declarative API: Offers a declarative API that builds a responsive and better touch-based experience.
Built-in Handler: Provides built-in gesture handlers that can understand users' common gestures, such as taps, swipes, rotation, long presses, and pan gestures. It aims to replace React Native's built-in Gesture Responder System, which has performance limitations as it runs on the JavaScript thread. So, every time a gesture event is carried out, the app has to go through several phases to determine the user's intention. This can lead to a more extended response of a touch. React Native Gesture Handler can eliminate that and give a seamless experience. 6. React Native UI Kitten It is a free and open-source library based on the Eva Design systems, which give the app a clean visual experience. Domain Variation: This can be a starter kit for building mobile applications in various domains such as e-commerce, social media, fitness, and more.
Dark and Light theme: UI Kitten creates appealing UIs with 20+ customizable components with light and dark themes, saving you time on design details.
No reloading: Allows runtime themes to instantly switch between different visual styles without reloading the application or modifying its source code. Easy to use: Once you install the UI Kitten package, you can import and use the components in your project. 7. React Native Gifted Chat Accessible communication is an essential aspect of user experience. Users should be allowed to communicate efficiently within the app. React Native Gifted Chat is one such library that provides chat functionality in the app. Built-in Components: Offers inbuilt components like bubbles, avatar views, input text fields, text messaging, emojis, and more.
Personalization: Comes with available option of further customizing them to match the brand identity and the overall look and feel of the app.
Multimedia Support: This library supports multimedia content like Images, videos, and audio messaging that can be integrated into the chat interface.
Customization: Users can customize their chat interface by changing the color, font, and avatar style. Conclusion These libraries offer various customization options, themes, and components that can give React native apps a clean and appealing user interface. Using these libraries, developers can create a one-of-a-kind experience for users, boosting retention and conversions. It also reduces development time and costs but needs effective deployment. This is where a seasoned  React native developer can help. The best way is to hire React native developers and leverage these libraries for optimal performance at low cost and time. React Native has been a hot topic among developers since its launch by Meta in 2015. Want to know the reason? It’s pretty simple—“Ultimate UX.” Yes, it offers UX that you, as a developer or entrepreneur, have never experienced. But your users can! HOW? I’ll tell you all about it and something more about React Native, which can help you achieve the ultimate UX with low cost and effort. Let's begin by understanding why it’s such a magnet attracting app developers worldwide. This framework allowed developers to build apps using a single code across platforms. This saved time and money by reducing the need to create separate codes for each platform. You might be skeptical about the quality of the apps in such cases. What if I tell you there is a way to ensure quality, optimize cost, and provide enhanced UX? Does that seem far from the truth? Well, not really! I recently came across React Native Libraries, which are easy to set up, take less time, and save money. Want to know which one of them is best for you? Here is a list! Top 7 React Native Libraries For Ultimate UX! Here are the top Libraries that will help you build a React Native App with a great user experience: 1. React Native Elements React Native Elements is a UI component library for building apps with better navigation and screen space. It has over 24,700 stars on GitHub. React Native Elements Navigation components: Gets components like badges and avatars, headers and tabs, and features for handling user interactions like checkboxes and switches. Screen space: This feature provides tools for creating dialogues and bottom sheets, allowing for additional content without taking up too much screen space. Community and updates: Its active and regular community ensures the app stays updated with the latest design trends and best practices. Navigation components: Gets components like badges and avatars, headers and tabs, and features for handling user interactions like checkboxes and switches. Navigation components: Screen space: This feature provides tools for creating dialogues and bottom sheets, allowing for additional content without taking up too much screen space. Screen space: Community and updates: Its active and regular community ensures the app stays updated with the latest design trends and best practices. Community and updates: 2. React Native Paper React Native Paper gets over 10,000 stars on GitHub. It focuses on providing a consistent look for your app across all platforms (iOS and Android). The best part, it follows Google's Material Design guidelines . React Native Paper Material Design guidelines Now, for those unfamiliar with material design let me explain in brief. Material design refers to print design methods that include typography, grids, space, scale, color, and imagery. It helps to create a hierarchy, meaning, and focus that gives users an immersive experience. Material design style: Employs Material design styles and themes, ensuring that apps developed using React Native Paper have a clean and professional appearance. Material design style: Employs Material design styles and themes, ensuring that apps developed using React Native Paper have a clean and professional appearance. Material design style: Dark Mode: Supports dark mode, which can significantly enhance the user experience, especially in low-light environments. Customization: Customizable style of components to match the specific needs of the brand identity while maintaining consistency. Customization: Customizable style of components to match the specific needs of the brand identity while maintaining consistency. Customization: API Integration: Technically, React Native Paper leverages native platform APIs for optimal performance and a natural feel. API Integration: 3. React Native Vector Icons React Native Vector Icons specializes in icons and boasts over 3,000 appealing and scalable icons that can be used in your React Native app. It has over 15,000 starts on GitHub. React Native Vector Icons Icons Collection: The library bundles a massive collection of icons from popular icon sets, such as FontAwesome, Entypo, ion icons, Material Icons, and more. Crisp rendering: These scalable icons extend, style, and integrate seamlessly into your project, ensuring crisp rendering across all screen resolutions without quality loss. Flexibility: These icons are versatile, functioning as emojis alongside text, button components, or native image elements. SVG Support: It supports SVG via Fontello or regular icon fonts, making it easy to customize the appearance of the icons. Time saver: The library is handy for saving time by not having to create custom icons from scratch. Icons Collection: The library bundles a massive collection of icons from popular icon sets, such as FontAwesome, Entypo, ion icons, Material Icons, and more. Icons Collection: collection of icons Crisp rendering: These scalable icons extend, style, and integrate seamlessly into your project, ensuring crisp rendering across all screen resolutions without quality loss. Crisp rendering: Flexibility: These icons are versatile, functioning as emojis alongside text, button components, or native image elements. Flexibility: SVG Support: It supports SVG via Fontello or regular icon fonts , making it easy to customize the appearance of the icons. SVG Support: regular icon fonts Time saver: The library is handy for saving time by not having to create custom icons from scratch. Time saver: 4. React Native Navigation React Native Navigation is a library that allows the creation of complex navigation flows easily and provides features like: React Native Navigation Stack navigation (used for hierarchical navigation), Drawer navigation (for side-menu navigation), Tab navigators (for bottom tab navigation) and Modal navigators (for overlays). Stack navigation (used for hierarchical navigation), Drawer navigation (for side-menu navigation), Tab navigators (for bottom tab navigation) and Modal navigators (for overlays). Other crucial features are, Animation and Transitions: Offers customizable navigation features like animations and transitions, and it uses an API to manage them and the user journey as well. Supports Deep Linking: Allows users to navigate specific parts of your app directly through external links. Animation and Transitions: Offers customizable navigation features like animations and transitions, and it uses an API to manage them and the user journey as well. Animation and Transitions: API Supports Deep Linking: Allows users to navigate specific parts of your app directly through external links. Supports Deep Linking: 5. React Native Gesture Handler Users are increasingly looking for gesture-based interaction in mobile apps, and this library is specially designed for it. React Native Gesture Handler lets developers create responsive gesture-based interactions in their React Native apps. React Native Gesture Handler Declarative API: Offers a declarative API that builds a responsive and better touch-based experience. Built-in Handler: Provides built-in gesture handlers that can understand users' common gestures, such as taps, swipes, rotation, long presses, and pan gestures. Declarative API: Offers a declarative API that builds a responsive and better touch-based experience. Declarative API: Built-in Handler: Provides built-in gesture handlers that can understand users' common gestures, such as taps, swipes, rotation, long presses, and pan gestures. Built-in Handler: It aims to replace React Native's built-in Gesture Responder System, which has performance limitations as it runs on the JavaScript thread. So, every time a gesture event is carried out, the app has to go through several phases to determine the user's intention. This can lead to a more extended response of a touch. React Native Gesture Handler can eliminate that and give a seamless experience. 6. React Native UI Kitten It is a free and open-source library based on the Eva Design systems, which give the app a clean visual experience. open-source library Domain Variation: This can be a starter kit for building mobile applications in various domains such as e-commerce, social media, fitness, and more. Dark and Light theme: UI Kitten creates appealing UIs with 20+ customizable components with light and dark themes, saving you time on design details. No reloading: Allows runtime themes to instantly switch between different visual styles without reloading the application or modifying its source code. Domain Variation: This can be a starter kit for building mobile applications in various domains such as e-commerce, social media, fitness, and more. Domain Variation: Dark and Light theme: UI Kitten creates appealing UIs with 20+ customizable components with light and dark themes, saving you time on design details. Dark and Light theme: No reloading: Allows runtime themes to instantly switch between different visual styles without reloading the application or modifying its source code. No reloading: Easy to use: Once you install the UI Kitten package, you can import and use the components in your project. 7. React Native Gifted Chat Accessible communication is an essential aspect of user experience. Users should be allowed to communicate efficiently within the app. React Native Gifted Chat is one such library that provides chat functionality in the app. React Native Gifted Chat Built-in Components: Offers inbuilt components like bubbles, avatar views, input text fields, text messaging, emojis, and more. Personalization: Comes with available option of further customizing them to match the brand identity and the overall look and feel of the app. Multimedia Support: This library supports multimedia content like Images, videos, and audio messaging that can be integrated into the chat interface. Customization: Users can customize their chat interface by changing the color, font, and avatar style. Built-in Components: Offers inbuilt components like bubbles, avatar views, input text fields, text messaging, emojis, and more. Built-in Components: Personalization: Comes with available option of further customizing them to match the brand identity and the overall look and feel of the app. Personalization: Multimedia Support: This library supports multimedia content like Images, videos, and audio messaging that can be integrated into the chat interface. Multimedia Support: Customization: Users can customize their chat interface by changing the color, font, and avatar style. Customization: Conclusion These libraries offer various customization options, themes, and components that can give React native apps a clean and appealing user interface. Using these libraries, developers can create a one-of-a-kind experience for users, boosting retention and conversions. It also reduces development time and costs but needs effective deployment. This is where a seasoned  React native developer can help. The best way is to hire React native developers and leverage these libraries for optimal performance at low cost and time.

7 React Native Libraries For Ultimate UX You Never Experienced

Fine-Tuning LLaMA for Multi-Stage Text Retrieval: Experiments

Authors:
(1) Busra Tabak [0000 −0001 −7460 −3689], Bogazici University, Turkey {busra.tabak@boun.edu.tr};
(2) Fatma Basak Aydemir [0000 −0003 −3833 −3997],  Bogazici University, Turkey {basak.aydemir@boun.edu.tr}. Table of Links Abstract and Introduction
Background
Approach
Experiments and Results
Discussion and Qualitative Analysis
Related Work
Conclusions and Future Work, and References 4 Experiments and Results This section presents the findings and outcomes of the conducted experiments, providing a comprehensive analysis of the collected data. We critically assess the performance and effectiveness of the proposed methodology by employing various evaluation metrics and techniques. We compare the extracted features from different perspectives, evaluating their individual contributions to the classification task. Lastly, we present a thorough statistical examination of the obtained results, employing appropriate statistical tests and measures to validate the significance and reliability of the findings. 4.1 Evaluation Table 5 presents the experiment results for the issue assignment. The models are evaluated on Team Assignment (TA) and Developer Assignment (DA). The stacking model (RF and Linear SVC) achieved the highest accuracy, with values of 0.92 for TA and 0.89 for DA. Other models, such as Support Vector Machine, Logistic Regression, and Random Forest, also showed good performance with accuracies ranging from 0.86 to 0.88. The transformer-based models, including DistilBert, Roberta, and Electra, demonstrated competitive accuracies, with Roberta and Electra achieving the highest scores in some cases. Table 6 provides the performance metrics for each class in the Stacking algorithm for issue assignment. Under the Team Assignment (TA) approach, the Stacking algorithm achieved a high precision value of 0.90 for the Developer class, indicating a low rate of false positive assignments. The Recall score of 0.99 for the Developer class demonstrates the algorithm’s ability to correctly identify the majority of instances assigned to developers. The Tester class shows a balance between precision (0.95) and recall (0.71), indicating accurate assignments with a relatively high rate of false negatives. The Designer class exhibits similar trends with a precision of 0.92 and a recall of 0.67. The Leader class has relatively lower precision and recall scores, indicating more challenging assignments for the algorithm. Under the Developer Assignment (DA) approach, the Stacking algorithm achieved high precision values for the Senior class (0.90) and the Mid class (0.89), indicating accurate assignments with low rates of false positives. The Mid class also demonstrates a high recall score of 0.94, indicating effective identification of instances assigned to this class. The Junior class shows a lower precision (0.62) and recall (0.53) compared to the other classes, suggesting potential challenges in accurately assigning instances to this class. We also test our classification algorithms using the most popular word embedding techniques to determine how well our features work. Figure 5 illustrates the comparison of our feature set with Tf-Idf and BOW methods for the issue assignment. Despite the potential of Word2Vec as a word embedding algorithm, the accuracy results in my approach do not yield comparable outcomes. We use the accuracy score as the comparison metric. The graph demonstrates that using our feature set yields superior results while using Tf-Idf and BOW yields comparable results. 4.2 Feature Comparison Reducing the number of redundant and irrelevant features is an effective way to improve the running time and generalization capability of a learning algorithm [12]. Feature selection methods are used to choose a subset of relevant features that contribute to the intended concept. These methods can employ a variety of models and techniques to calculate feature importance scores. One simple approach [9] is to calculate the coefficient statistics between each feature and the target variable. This method can help to identify the most important features of a given problem and discard the redundant or irrelevant ones. By reducing the number of features used for training a model, the running time of the algorithm can be significantly reduced without sacrificing accuracy. Moreover, feature selection can also improve the interpretability of the model, as it helps to identify the key factors that influence the target variable. We present the coefficient values of each feature in Figure 6. We find that Issue Type, namely FJC1, emerges as the most influential feature from our feature set. Apart from the Issue Type, we discover that the features Watchers and Summary Words also exhibit significant effectiveness in our analysis. Conversely, features such as Reopen Count, Test Execution Type, and Approval Type demonstrate no impact on our issue assignment process. In Figure 7, we present the effective POS tags in Turkish, highlighting the most influential ones among all POS tags. Notably, the number of unknown words, verbs, and nouns emerge as the most impactful features. Following the rigorous selection of the best features, we proceed to employ Scikit-learn’s [38] SelectFromModel, a powerful meta-transformer designed to choose features based on their importance weights, to retrain our models. Through this process, we carefully identify and select the top eight features that exhibited the highest significance, as determined by the module. Remarkably, leveraging this refined feature subset allows us to achieve optimal performance and attain the most favorable outcome in our experiments. 4.3 Statistical Analysis Statistical significance tests are conducted to compare classification methods for determining whether one learning algorithm outperforms another on a particular learning task. Dietterich [14] reviews five approximate statistical tests and concludes that McNemar’s test and the 5x2 cv t-test, both have low type I error and sufficient power. In our study, we combine all data sets into a single data set for the classification algorithm. Dietterich [14] recommends using a 5x2 ttest to statistically compare two classifiers on a single data set. The 5x2 f-test, which is also suggested as the new standard by the original authors above, is further expanded upon by Alpaydın [4]. The following lists the null hypothesis and the alternative hypothesis. The null hypothesis is that the probabilities are the same, or in simpler terms, neither of the two models outperforms the other. The alternative hypothesis, therefore, holds that the performances of the two models are not equivalent. Accordingly, we apply the 5x2 f-test implemented by Alpaydın [4] which is an extension of the 5x2 cv t-test as stated above. We create the matrix for all pairwise comparisons of learning algorithms. In this test, the splitting process (50% training and 50% test data) is repeated five times. A and B are fitted to the training split and their performance on the test split in each of the five iterations is assessed. The training and test sets are then rotated (the training set becomes the test set, and vice versa), and the performance is computed again, yielding two performance difference measures. Then, the mean and variance of the differences are estimated and the f-statistic proposed by Alpaydın is calculated as This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. Authors: (1) Busra Tabak [0000 −0001 −7460 −3689], Bogazici University, Turkey {busra.tabak@boun.edu.tr}; (2) Fatma Basak Aydemir [0000 −0003 −3833 −3997],  Bogazici University, Turkey {basak.aydemir@boun.edu.tr}. Authors: Authors: (1) Busra Tabak [0000 −0001 −7460 −3689], Bogazici University, Turkey {busra.tabak@boun.edu.tr}; (2) Fatma Basak Aydemir [0000 −0003 −3833 −3997],  Bogazici University, Turkey {basak.aydemir@boun.edu.tr}. Table of Links Abstract and Introduction Background Approach Experiments and Results Discussion and Qualitative Analysis Related Work Conclusions and Future Work, and References Abstract and Introduction Abstract and Introduction Background Background Approach Approach Experiments and Results Experiments and Results Discussion and Qualitative Analysis Discussion and Qualitative Analysis Related Work Related Work Conclusions and Future Work, and References Conclusions and Future Work, and References 4 Experiments and Results This section presents the findings and outcomes of the conducted experiments, providing a comprehensive analysis of the collected data. We critically assess the performance and effectiveness of the proposed methodology by employing various evaluation metrics and techniques. We compare the extracted features from different perspectives, evaluating their individual contributions to the classification task. Lastly, we present a thorough statistical examination of the obtained results, employing appropriate statistical tests and measures to validate the significance and reliability of the findings. 4.1 Evaluation Table 5 presents the experiment results for the issue assignment. The models are evaluated on Team Assignment (TA) and Developer Assignment (DA). The stacking model (RF and Linear SVC) achieved the highest accuracy, with values of 0.92 for TA and 0.89 for DA. Other models, such as Support Vector Machine, Logistic Regression, and Random Forest, also showed good performance with accuracies ranging from 0.86 to 0.88. The transformer-based models, including DistilBert, Roberta, and Electra, demonstrated competitive accuracies, with Roberta and Electra achieving the highest scores in some cases. Table 6 provides the performance metrics for each class in the Stacking algorithm for issue assignment. Under the Team Assignment (TA) approach, the Stacking algorithm achieved a high precision value of 0.90 for the Developer class, indicating a low rate of false positive assignments. The Recall score of 0.99 for the Developer class demonstrates the algorithm’s ability to correctly identify the majority of instances assigned to developers. The Tester class shows a balance between precision (0.95) and recall (0.71), indicating accurate assignments with a relatively high rate of false negatives. The Designer class exhibits similar trends with a precision of 0.92 and a recall of 0.67. The Leader class has relatively lower precision and recall scores, indicating more challenging assignments for the algorithm. Under the Developer Assignment (DA) approach, the Stacking algorithm achieved high precision values for the Senior class (0.90) and the Mid class (0.89), indicating accurate assignments with low rates of false positives. The Mid class also demonstrates a high recall score of 0.94, indicating effective identification of instances assigned to this class. The Junior class shows a lower precision (0.62) and recall (0.53) compared to the other classes, suggesting potential challenges in accurately assigning instances to this class. We also test our classification algorithms using the most popular word embedding techniques to determine how well our features work. Figure 5 illustrates the comparison of our feature set with Tf-Idf and BOW methods for the issue assignment. Despite the potential of Word2Vec as a word embedding algorithm, the accuracy results in my approach do not yield comparable outcomes. We use the accuracy score as the comparison metric. The graph demonstrates that using our feature set yields superior results while using Tf-Idf and BOW yields comparable results. 4.2 Feature Comparison Reducing the number of redundant and irrelevant features is an effective way to improve the running time and generalization capability of a learning algorithm [12]. Feature selection methods are used to choose a subset of relevant features that contribute to the intended concept. These methods can employ a variety of models and techniques to calculate feature importance scores. One simple approach [9] is to calculate the coefficient statistics between each feature and the target variable. This method can help to identify the most important features of a given problem and discard the redundant or irrelevant ones. By reducing the number of features used for training a model, the running time of the algorithm can be significantly reduced without sacrificing accuracy. Moreover, feature selection can also improve the interpretability of the model, as it helps to identify the key factors that influence the target variable. We present the coefficient values of each feature in Figure 6. We find that Issue Type, namely FJC1, emerges as the most influential feature from our feature set. Apart from the Issue Type, we discover that the features Watchers and Summary Words also exhibit significant effectiveness in our analysis. Conversely, features such as Reopen Count, Test Execution Type, and Approval Type demonstrate no impact on our issue assignment process. In Figure 7, we present the effective POS tags in Turkish, highlighting the most influential ones among all POS tags. Notably, the number of unknown words, verbs, and nouns emerge as the most impactful features. Following the rigorous selection of the best features, we proceed to employ Scikit-learn’s [38] SelectFromModel, a powerful meta-transformer designed to choose features based on their importance weights, to retrain our models. Through this process, we carefully identify and select the top eight features that exhibited the highest significance, as determined by the module. Remarkably, leveraging this refined feature subset allows us to achieve optimal performance and attain the most favorable outcome in our experiments. 4.3 Statistical Analysis Statistical significance tests are conducted to compare classification methods for determining whether one learning algorithm outperforms another on a particular learning task. Dietterich [14] reviews five approximate statistical tests and concludes that McNemar’s test and the 5x2 cv t-test, both have low type I error and sufficient power. In our study, we combine all data sets into a single data set for the classification algorithm. Dietterich [14] recommends using a 5x2 ttest to statistically compare two classifiers on a single data set. The 5x2 f-test, which is also suggested as the new standard by the original authors above, is further expanded upon by Alpaydın [4]. The following lists the null hypothesis and the alternative hypothesis. The null hypothesis is that the probabilities are the same, or in simpler terms, neither of the two models outperforms the other. The alternative hypothesis, therefore, holds that the performances of the two models are not equivalent. Accordingly, we apply the 5x2 f-test implemented by Alpaydın [4] which is an extension of the 5x2 cv t-test as stated above. We create the matrix for all pairwise comparisons of learning algorithms. In this test, the splitting process (50% training and 50% test data) is repeated five times. A and B are fitted to the training split and their performance on the test split in each of the five iterations is assessed. The training and test sets are then rotated (the training set becomes the test set, and vice versa), and the performance is computed again, yielding two performance difference measures. Then, the mean and variance of the differences are estimated and the f-statistic proposed by Alpaydın is calculated as This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. available on arxiv

Comparison of Machine Learning Methods: Experiments and Results

Authors:
(1) Lewis Tunstall,  Equal contribution and The H4 (Helpful, Honest, Harmless, Huggy) Team (email: lewis@huggingface.co);
(2) Edward Beeching, Equal contribution and The H4 (Helpful, Honest, Harmless, Huggy) Team;
(3) Nathan Lambert, The H4 (Helpful, Honest, Harmless, Huggy) Team;
(4) Nazneen Rajani, The H4 (Helpful, Honest, Harmless, Huggy) Team;
(5) Kashif Rasul, The H4 (Helpful, Honest, Harmless, Huggy) Team;
(6) Younes Belkada, The H4 (Helpful, Honest, Harmless, Huggy) Team;
(7) Shengyi Huang, The H4 (Helpful, Honest, Harmless, Huggy) Team;
(8) Leandro von Werra, The H4 (Helpful, Honest, Harmless, Huggy) Team;
(9) Clementine Fourrier, The H4 (Helpful, Honest, Harmless, Huggy) Team;
(10) Nathan Habib, The H4 (Helpful, Honest, Harmless, Huggy) Team;
(11) Nathan Sarrazin, The H4 (Helpful, Honest, Harmless, Huggy) Team;
(12) Omar Sanseviero, The H4 (Helpful, Honest, Harmless, Huggy) Team;
(13) Alexander M. Rush, The H4 (Helpful, Honest, Harmless, Huggy) Team;
(14) Thomas Wolf, The H4 (Helpful, Honest, Harmless, Huggy) Team. Table of Links Abstract and Introduction
Related Work
Method
Experimental Details
Results and Ablations
Conclusions and Limitations , Acknowledgements and References
Appendix 4 EXPERIMENTAL DETAILS We conduct all of our fine-tuning experiments using Mistral 7B (Jiang et al., 2023), which is the current state-of-the-art base LM at the 7B parameter scale, and matches the performance of much larger models like LLaMa 34B on many NLP benchmarks. We use the Transformer Reinforcement Learning (TRL) library for fine-tuning (von Werra et al., 2020), in conjunction with DeepSpeed ZeRO3 (Rajbhandari et al., 2020) and FlashAttention-2 (Dao, 2023) to optimize memory and improve training speed. All models are trained with the AdamW optimizer and no weight decay. We did not experiment with parameter-efficient techniques such as LoRA (Hu et al., 2021), but expect similar results to hold with these methods. All experiments were run on 16 A100s using bfloat16 precision and typically took 2-4 hours to complete. For the full set of hyperparameters and instructions on how to train the models, see: https://github.com/huggingface/alignment-handbook. 4.1 DATASETS We focus on two dialogue datasets that have been distilled from a mix of open and proprietary models, and have previously been shown to produce strong chat models like the UltraLM (Ding et al., 2023): • UltraChat (Ding et al., 2023) is a self-refinement dataset consisting of 1.47M multi-turn dialogues generated by GPT-3.5-TURBO over 30 topics and 20 different types of text material. We initially ran dSFT over the whole corpus, but found the resulting chat model had a tendency to respond with incorrect capitalization and would preface its answers with phrases such as “I don’t have personal experiences”, even for straightforward questions like “How do I clean my car?”. To handle these issues in the training data, we applied truecasing heuristics to fix the grammatical errors (approximately 5% of the dataset), as well as several filters to focus on helpfulness and remove the undesired model responses. The resulting dataset contains approximately 200k examples. • UltraFeedback (Cui et al., 2023) consists of 64k prompts, each of which have four LLM responses that are rated by GPT-4 according to criteria like instruction-following, honesty, and helpfulness. We construct binary preferences from UltraFeedback by selecting the highest mean score as the “chosen” response and one of the remaining three at random as “rejected”. We opted for random selection instead of selecting the lowest-scored response to encourage diversity and make the DPO objective more challenging. As noted above, this step is computed offline and does not involve any sampling from the reference model. We make the pre-processed datasets available on the Hugging Face Hub.1 4.2 EVALUATION Our main evaluations are on single-turn and multi-turn chat benchmarks that measure a model’s ability to follow instructions and respond to challenging prompts across a diverse range of domains: • MT-Bench (Zheng et al., 2023) is a multi-turn benchmark that consists of 160 questions across eight different areas of knowledge. In this benchmark, the model must answer an initial question, and then provide a second response to a predefined followup question. Each model response is then rated by GPT-4 on a scale from 1-10, with the final score given by the mean over the two turns. • AlpacaEval (Li et al., 2023) is a single-turn benchmark where a model must generate a response to 805 questions on different topics, mostly focused on helpfulness. Models are also scored by GPT-4, but the final metric is the pairwise win-rate against a baseline model (text-davinci-003). We also evaluate ZEPHYR-7B on the Open LLM Leaderboard (Beeching et al., 2023), which measures the performance of LMs across four multiclass classification tasks: ARC (Clark et al., 2018), HellaSwag (Zellers et al., 2019), MMLU (Hendrycks et al., 2021), and Truthful QA(Lin et al., 2022). Although this leaderboard does not directly measure the conversational quality of chat models, it does provide a useful signal to validate whether fine-tuning has introduced regressions on the base model’s reasoning and truthfulness capabilities. Across all benchmarks, we compare ZEPHYR-7B against a variety of open and proprietary models, each with different alignment procedures. To facilitate comparison across open model sizes, we group our comparisons in terms of 7B models (XWIN-LM (Team, 2023), MISTRALINSTRUCT (Jiang et al., 2023), MPT-CHAT (ML, 2023), and STABLELM-α), as well as larger models up to 70B parameters (LLAMA2-CHAT (Touvron et al., 2023), VICUNA˜ (Chiang et al., 2023), WizardLM (Xu et al.), and GUANACO (Dettmers et al., 2023)). For the chat benchmarks, we also compare against proprietary models, including CLAUDE 2, GPT-3.5-TURBO and GPT-4 (OpenAI, 2023). 4.3 DETAILS OF SFT TRAINING We train our SFT models for one to three epochs. We use a cosine learning rate scheduler with a peak learning rate of 2e-5 and 10% warmup steps. We train all models with a global batch size of 512 and use packing with a sequence length of 2048 tokens. 4.4 DETAILS OF DPO TRAINING Similar to SFT, we train our DPO models for one to three epochs. We use a linear learning rate scheduler with a peak learning rate of 5e-7 and 10% warmup steps. We train all models with a global batch size of 32 and use β = 0.1 from Eq. (1) to control the deviation from the reference model. The final ZEPHYR-7B model was initialized from the SFT model that was trained for one epoch and further optimized for three DPO epochs (see Figure 3 for an epoch ablation on MT-Bench). This paper is available on arxiv under CC 4.0 license. 1 https://huggingface.co/collections/HuggingFaceH4/ zephyr-7b-6538c6d6d5ddd1cbb1744a66 Authors: (1) Lewis Tunstall,  Equal contribution and The H4 (Helpful, Honest, Harmless, Huggy) Team (email: lewis@huggingface.co); (2) Edward Beeching, Equal contribution and The H4 (Helpful, Honest, Harmless, Huggy) Team; (3) Nathan Lambert, The H4 (Helpful, Honest, Harmless, Huggy) Team; (4) Nazneen Rajani, The H4 (Helpful, Honest, Harmless, Huggy) Team; (5) Kashif Rasul, The H4 (Helpful, Honest, Harmless, Huggy) Team; (6) Younes Belkada, The H4 (Helpful, Honest, Harmless, Huggy) Team; (7) Shengyi Huang, The H4 (Helpful, Honest, Harmless, Huggy) Team; (8) Leandro von Werra, The H4 (Helpful, Honest, Harmless, Huggy) Team; (9) Clementine Fourrier, The H4 (Helpful, Honest, Harmless, Huggy) Team; (10) Nathan Habib, The H4 (Helpful, Honest, Harmless, Huggy) Team; (11) Nathan Sarrazin, The H4 (Helpful, Honest, Harmless, Huggy) Team; (12) Omar Sanseviero, The H4 (Helpful, Honest, Harmless, Huggy) Team; (13) Alexander M. Rush, The H4 (Helpful, Honest, Harmless, Huggy) Team; (14) Thomas Wolf, The H4 (Helpful, Honest, Harmless, Huggy) Team. Authors: Authors: (1) Lewis Tunstall,  Equal contribution and The H4 (Helpful, Honest, Harmless, Huggy) Team (email: lewis@huggingface.co); (2) Edward Beeching, Equal contribution and The H4 (Helpful, Honest, Harmless, Huggy) Team; (3) Nathan Lambert, The H4 (Helpful, Honest, Harmless, Huggy) Team; (4) Nazneen Rajani, The H4 (Helpful, Honest, Harmless, Huggy) Team; (5) Kashif Rasul, The H4 (Helpful, Honest, Harmless, Huggy) Team; (6) Younes Belkada, The H4 (Helpful, Honest, Harmless, Huggy) Team; (7) Shengyi Huang, The H4 (Helpful, Honest, Harmless, Huggy) Team; (8) Leandro von Werra, The H4 (Helpful, Honest, Harmless, Huggy) Team; (9) Clementine Fourrier, The H4 (Helpful, Honest, Harmless, Huggy) Team; (10) Nathan Habib, The H4 (Helpful, Honest, Harmless, Huggy) Team; (11) Nathan Sarrazin, The H4 (Helpful, Honest, Harmless, Huggy) Team; (12) Omar Sanseviero, The H4 (Helpful, Honest, Harmless, Huggy) Team; (13) Alexander M. Rush, The H4 (Helpful, Honest, Harmless, Huggy) Team; (14) Thomas Wolf, The H4 (Helpful, Honest, Harmless, Huggy) Team. Table of Links Abstract and Introduction Related Work Method Experimental Details Results and Ablations Conclusions and Limitations , Acknowledgements and References Appendix Abstract and Introduction Abstract and Introduction Related Work Related Work Method Method Experimental Details Experimental Details Results and Ablations Results and Ablations Conclusions and Limitations , Acknowledgements and References Conclusions and Limitations , Acknowledgements and References Appendix Appendix 4 EXPERIMENTAL DETAILS We conduct all of our fine-tuning experiments using Mistral 7B (Jiang et al., 2023), which is the current state-of-the-art base LM at the 7B parameter scale, and matches the performance of much larger models like LLaMa 34B on many NLP benchmarks. We use the Transformer Reinforcement Learning (TRL) library for fine-tuning (von Werra et al., 2020), in conjunction with DeepSpeed ZeRO3 (Rajbhandari et al., 2020) and FlashAttention-2 (Dao, 2023) to optimize memory and improve training speed. All models are trained with the AdamW optimizer and no weight decay. We did not experiment with parameter-efficient techniques such as LoRA (Hu et al., 2021), but expect similar results to hold with these methods. All experiments were run on 16 A100s using bfloat16 precision and typically took 2-4 hours to complete. For the full set of hyperparameters and instructions on how to train the models, see: https://github.com/huggingface/alignment-handbook. 4.1 DATASETS We focus on two dialogue datasets that have been distilled from a mix of open and proprietary models, and have previously been shown to produce strong chat models like the UltraLM (Ding et al., 2023): • UltraChat (Ding et al., 2023) is a self-refinement dataset consisting of 1.47M multi-turn dialogues generated by GPT-3.5-TURBO over 30 topics and 20 different types of text material. We initially ran dSFT over the whole corpus, but found the resulting chat model had a tendency to respond with incorrect capitalization and would preface its answers with phrases such as “I don’t have personal experiences”, even for straightforward questions like “How do I clean my car?”. To handle these issues in the training data, we applied truecasing heuristics to fix the grammatical errors (approximately 5% of the dataset), as well as several filters to focus on helpfulness and remove the undesired model responses. The resulting dataset contains approximately 200k examples. • UltraChat • UltraFeedback (Cui et al., 2023) consists of 64k prompts, each of which have four LLM responses that are rated by GPT-4 according to criteria like instruction-following, honesty, and helpfulness. We construct binary preferences from UltraFeedback by selecting the highest mean score as the “chosen” response and one of the remaining three at random as “rejected”. We opted for random selection instead of selecting the lowest-scored response to encourage diversity and make the DPO objective more challenging. As noted above, this step is computed offline and does not involve any sampling from the reference model. • UltraFeedback We make the pre-processed datasets available on the Hugging Face Hub.1 4.2 EVALUATION Our main evaluations are on single-turn and multi-turn chat benchmarks that measure a model’s ability to follow instructions and respond to challenging prompts across a diverse range of domains: • MT-Bench (Zheng et al., 2023) is a multi-turn benchmark that consists of 160 questions across eight different areas of knowledge. In this benchmark, the model must answer an initial question, and then provide a second response to a predefined followup question. Each model response is then rated by GPT-4 on a scale from 1-10, with the final score given by the mean over the two turns. • MT-Bench • AlpacaEval (Li et al., 2023) is a single-turn benchmark where a model must generate a response to 805 questions on different topics, mostly focused on helpfulness. Models are also scored by GPT-4, but the final metric is the pairwise win-rate against a baseline model (text-davinci-003). • AlpacaEval We also evaluate ZEPHYR-7B on the Open LLM Leaderboard (Beeching et al., 2023), which measures the performance of LMs across four multiclass classification tasks: ARC (Clark et al., 2018), HellaSwag (Zellers et al., 2019), MMLU (Hendrycks et al., 2021), and Truthful QA(Lin et al., 2022). Although this leaderboard does not directly measure the conversational quality of chat models, it does provide a useful signal to validate whether fine-tuning has introduced regressions on the base model’s reasoning and truthfulness capabilities. Across all benchmarks, we compare ZEPHYR-7B against a variety of open and proprietary models, each with different alignment procedures. To facilitate comparison across open model sizes, we group our comparisons in terms of 7B models (XWIN-LM (Team, 2023), MISTRALINSTRUCT (Jiang et al., 2023), MPT-CHAT (ML, 2023), and STABLELM-α), as well as larger models up to 70B parameters (LLAMA2-CHAT (Touvron et al., 2023), VICUNA˜ (Chiang et al., 2023), WizardLM (Xu et al.), and GUANACO (Dettmers et al., 2023)). For the chat benchmarks, we also compare against proprietary models, including CLAUDE 2, GPT-3.5-TURBO and GPT-4 (OpenAI, 2023). 4.3 DETAILS OF SFT TRAINING We train our SFT models for one to three epochs. We use a cosine learning rate scheduler with a peak learning rate of 2e-5 and 10% warmup steps. We train all models with a global batch size of 512 and use packing with a sequence length of 2048 tokens. 4.4 DETAILS OF DPO TRAINING Similar to SFT, we train our DPO models for one to three epochs. We use a linear learning rate scheduler with a peak learning rate of 5e-7 and 10% warmup steps. We train all models with a global batch size of 32 and use β = 0.1 from Eq. (1) to control the deviation from the reference model. The final ZEPHYR-7B model was initialized from the SFT model that was trained for one epoch and further optimized for three DPO epochs (see Figure 3 for an epoch ablation on MT-Bench). This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv 1 https://huggingface.co/collections/HuggingFaceH4/ zephyr-7b-6538c6d6d5ddd1cbb1744a66

Zephyr: Direct Distillation of LM Alignment: Experimental Details

The Noonification: The Odin Project: My 111-Day Experience With It (7/2/2024)

On January 22, 2024, I didn’t know how to write a line of code in JavaScript on my own. It was the day when I started my first lesson in the Foundation course by The Odin Project. This was the best thing I could find on my coding journey. The Odin Project saved me from tutorial hell, and all the projects you’ll see below were done on my own without any help from ChatGPT or other tools—just the knowledge I gained from The Odin Project. After exactly 440 hours and 20 minutes of learning over 111 days, from January 22, 2024, to May 11, 2024, I was able to complete the Foundations course and create the final project on my own, which was a calculator. Here are some additional statistics about the time I spent on this, as I know some people are curious about how long it takes. How can you copy the process? Let’s dive in. Reviewing the Data Let’s start by looking at this data to see how accurate it really is. I learned every single day for at least one hour. To calculate the study sessions, I used a Pomodoro timer set for 25 minutes each, with a 5-minute break in between. So, I assume that 2 Pomodoros equal a 1-hour study session. Tools I Used for Note-Taking What do I use to document everything? Obsidian notes. I began using it just before starting Odin because I knew this journey would be quite long, and I would need some nice notes to store my ideas. Then, actually, for fun, I decided to count all these hours to see how long it would really take. Many people asked, and only a few answered. It seems like most people just don’t bother to count the time, and I don’t blame them. (However, I saw some people claim it took them 3 months, while others needed a year). And I am talking only about the Foundations part here. My Study Routine But here’s how I did it, without skipping a single day, cheating, or omitting any resources. I read everything, sometimes including additional resources. This, however, depends on the information included or the blog quality. If I like it or see it would benefit me to save it as a bookmark for future use, I read it and save it; if not, I just skip and move further. But I would highly recommend you always open all additional resource links and at least poke around. For reference, you can have a look at all the projects I have done so far and get an idea of what you will be able to accomplish after the Foundation part of The Odin Project: RECIPES BOOK (yup, it’s ugly but as I remember, it was one of the first projects ;)
LANDING PAGE
ROCK-PAPER-SCISSORS
ETCH A SKETCH (this one wasn’t easy, believe me!)
CALCULATOR Weekdays My daily routine looks like this: I go to work from Monday to Friday, from 7:30 to 16:00. I’m home around 16:30, then I take a shower, eat a quick supper, and usually by 17:00, I’m ready to start. I spend one hour working on my other blogs that are not related to Odin. If sometimes it takes me only 30 minutes, then I start learning Odin at 17:30; if not, then at 18:00. I learn until 19:00 and then work out for about 30 minutes in my room. By 19:30, I go for a walk and come back around 21:00. Sometimes, I study for 15-30 minutes more, but not very often So usually, it’s 1 to 1.5 hours a day. I’d like to mention that I don’t have kids, so I don’t need to pick them up from school, etc. I live alone, which helps me manage distractions and stick to my plan. But even so, one hour is not a lot, and I think that everybody can do it. Weekends Weekends look a bit different though. I wake up around 6:30, do some stretching, and start working on Odin at 7:00. I am able to complete 4 Pomodoros by 10:00. Then I go tidy up my room, make breakfast, drink coffee, and am back to learning at 11:30. On Saturday, I take a rest from working out and study until 14:00, then go to make dinner. If it’s Sunday, however, I work out from 13:30 to 14:00, and then the rest looks the same as Saturday. After my walk, especially on Saturday, I go get groceries, and I’m back home around 17:00, which allows me to do 2-3 more Pomodoros. Altogether, I can complete around 20 Pomodoros during a whole weekend, which is like 10 hours. Sometimes it’s more, sometimes it’s less. If there are any days off, I treat them as if it were a weekend. The Power of the Pomodoro Technique A big part of my learning process was the Pomodoro Technique I already mentioned. It’s all about working in focused 25-minute bursts, called Pomodoros, with a 5-minute break in between. After four Pomodoros, you get a longer break of 15-30 minutes. This method helps you stay focused and avoid burnout. Barbara Oakley’s course on Coursera, “Learning How to Learn,” dives into how our brains work when we’re learning. She talks about the Pomodoro Technique as a great way to break study sessions into smaller chunks, making it easier for the brain to process and remember stuff. If you’re curious about the science behind effective learning, you should definitely check out this course. I think it’s still free. At least, it used to be when I was told about it when I was starting. The Odin Project’s curriculum uses similar principles. They suggest structured study sessions like Pomodoros, which align with proven learning methods. Following their directions and instructions helps you not only learn the material but also build strong study habits that will benefit you in the long run. Balancing Workouts and Learning If I didn’t work out, I would be able to learn more and faster, especially on weekdays, but it wouldn’t be too healthy, I bet. Tips for Staying Focused I also have some tips that worked out for me pretty well. Remember, you are going to face a lot of time-fighting procrastination between sessions. It’s very good to: Turn on flight mode on your phone, and put the device as far from you as you can. Don’t use your phone during the 5-minute break between Pomodoros. Stretch instead, walk around the room, and look through the window. If you are tired and feeling sleepy, get a standing desk or use drawers or some other furniture. Overcoming Challenges There will be days when you are tired or have many negative thoughts running through your head telling you to stop, that it’s not worth it, that you might be too stupid, or that AI will be coding in the future anyway. If all these happen during your Pomodoro sessions, DON’T WORRY AND DON’T GIVE UP. I also had loads of these thoughts, and it’s normal. Some days are worse than others, but I noticed that even the days when I was looking at the screen trying to read boring documentation 5 times helped me to at least stay consistent and build a habit. Finding Enjoyment in the Process At some point, you will realize that you like the process you are going through (if you haven’t liked it yet), and you will notice that all negative thoughts start to fade away, and your motivation goes up day after day, regardless of the level of difficulty. I wrote more about this in my blog post about the calculator project. Happy learning! What challenges have you faced in your learning journey? Share your best productivity tips in the comments! Follow me on Twitter for more tips and insights on coding and productivity. On January 22, 2024, I didn’t know how to write a line of code in JavaScript on my own. It was the day when I started my first lesson in the Foundation course by The Odin Project . This was the best thing I could find on my coding journey. The Odin Project saved me from tutorial hell , and all the projects you’ll see below were done on my own without any help from ChatGPT or other tools—just the knowledge I gained from The Odin Project. The Odin Project The Odin Project saved me from tutorial hell After exactly 440 hours and 20 minutes of learning over 111 days , from January 22, 2024, to May 11, 2024, I was able to complete the Foundations course and create the final project on my own, which was a calculator . 440 hours and 20 minutes 111 days calculator Here are some additional statistics about the time I spent on this, as I know some people are curious about how long it takes. How can you copy the process? Let’s dive in. Reviewing the Data Let’s start by looking at this data to see how accurate it really is. I learned every single day for at least one hour . To calculate the study sessions, I used a Pomodoro timer set for 25 minutes each, with a 5-minute break in between. every single day for at least one hour Pomodoro timer So, I assume that 2 Pomodoros equal a 1-hour study session. Tools I Used for Note-Taking What do I use to document everything? Obsidian notes . Obsidian notes I began using it just before starting Odin because I knew this journey would be quite long, and I would need some nice notes to store my ideas. Then, actually, for fun, I decided to count all these hours to see how long it would really take. Many people asked, and only a few answered. It seems like most people just don’t bother to count the time, and I don’t blame them. (However, I saw some people claim it took them 3 months, while others needed a year). And I am talking only about the Foundations part here. My Study Routine But here’s how I did it, without skipping a single day, cheating, or omitting any resources . without skipping a single day, cheating, or omitting any resources I read everything, sometimes including additional resources. This, however, depends on the information included or the blog quality. If I like it or see it would benefit me to save it as a bookmark for future use, I read it and save it; if not, I just skip and move further. But I would highly recommend you always open all additional resource links and at least poke around. always open all additional resource links For reference, you can have a look at all the projects I have done so far and get an idea of what you will be able to accomplish after the Foundation part of The Odin Project: RECIPES BOOK (yup, it’s ugly but as I remember, it was one of the first projects ;) LANDING PAGE ROCK-PAPER-SCISSORS ETCH A SKETCH (this one wasn’t easy, believe me!) CALCULATOR RECIPES BOOK (yup, it’s ugly but as I remember, it was one of the first projects ;) RECIPES BOOK LANDING PAGE LANDING PAGE ROCK-PAPER-SCISSORS ROCK-PAPER-SCISSORS ETCH A SKETCH (this one wasn’t easy, believe me!) ETCH A SKETCH CALCULATOR CALCULATOR Weekdays My daily routine looks like this: I go to work from Monday to Friday, from 7:30 to 16:00. I’m home around 16:30, then I take a shower, eat a quick supper, and usually by 17:00, I’m ready to start. I spend one hour working on my other blogs that are not related to Odin. If sometimes it takes me only 30 minutes, then I start learning Odin at 17:30; if not, then at 18:00. I learn until 19:00 and then work out for about 30 minutes in my room. By 19:30, I go for a walk and come back around 21:00. Sometimes, I study for 15-30 minutes more, but not very often So usually, it’s 1 to 1.5 hours a day. I’d like to mention that I don’t have kids , so I don’t need to pick them up from school, etc. I live alone , which helps me manage distractions and stick to my plan. I don’t have kids I live alone But even so, one hour is not a lot, and I think that everybody can do it. Weekends Weekends look a bit different though. I wake up around 6:30, do some stretching, and start working on Odin at 7:00. I am able to complete 4 Pomodoros by 10:00. Then I go tidy up my room, make breakfast, drink coffee, and am back to learning at 11:30. On Saturday, I take a rest from working out and study until 14:00, then go to make dinner. If it’s Sunday, however, I work out from 13:30 to 14:00, and then the rest looks the same as Saturday. After my walk, especially on Saturday, I go get groceries, and I’m back home around 17:00, which allows me to do 2-3 more Pomodoros. Altogether, I can complete around 20 Pomodoros during a whole weekend, which is like 10 hours. Sometimes it’s more, sometimes it’s less. If there are any days off, I treat them as if it were a weekend. The Power of the Pomodoro Technique A big part of my learning process was the Pomodoro Technique I already mentioned. It’s all about working in focused 25-minute bursts, called Pomodoros, with a 5-minute break in between. After four Pomodoros, you get a longer break of 15-30 minutes. This method helps you stay focused and avoid burnout. Barbara Oakley’s course on Coursera, “Learning How to Learn,” dives into how our brains work when we’re learning. She talks about the Pomodoro Technique as a great way to break study sessions into smaller chunks, making it easier for the brain to process and remember stuff. Barbara Oakley’s course on Coursera, “Learning How to Learn,” If you’re curious about the science behind effective learning, you should definitely check out this course. I think it’s still free . At least, it used to be when I was told about it when I was starting. I think it’s still free The Odin Project’s curriculum uses similar principles. They suggest structured study sessions like Pomodoros, which align with proven learning methods. Following their directions and instructions helps you not only learn the material but also build strong study habits that will benefit you in the long run . The Odin Project’s build strong study habits that will benefit you in the long run Balancing Workouts and Learning If I didn’t work out, I would be able to learn more and faster, especially on weekdays, but it wouldn’t be too healthy, I bet. Tips for Staying Focused I also have some tips that worked out for me pretty well. Remember, you are going to face a lot of time-fighting procrastination between sessions. It’s very good to: Turn on flight mode on your phone, and put the device as far from you as you can. Turn on flight mode on your phone, and put the device as far from you as you can. Turn on flight mode Don’t use your phone during the 5-minute break between Pomodoros. Stretch instead, walk around the room, and look through the window. Don’t use your phone during the 5-minute break between Pomodoros. Stretch instead, walk around the room, and look through the window. Don’t use your phone If you are tired and feeling sleepy, get a standing desk or use drawers or some other furniture. If you are tired and feeling sleepy, get a standing desk or use drawers or some other furniture. get a standing desk Overcoming Challenges There will be days when you are tired or have many negative thoughts running through your head telling you to stop, that it’s not worth it, that you might be too stupid, or that AI will be coding in the future anyway. If all these happen during your Pomodoro sessions, DON’T WORRY AND DON’T GIVE UP . DON’T WORRY AND DON’T GIVE UP I also had loads of these thoughts, and it’s normal. Some days are worse than others, but I noticed that even the days when I was looking at the screen trying to read boring documentation 5 times helped me to at least stay consistent and build a habit. Finding Enjoyment in the Process At some point, you will realize that you like the process you are going through (if you haven’t liked it yet), and you will notice that all negative thoughts start to fade away, and your motivation goes up day after day, regardless of the level of difficulty. I wrote more about this in my blog post about the calculator project . Happy learning! the calculator project What challenges have you faced in your learning journey? Share your best productivity tips in the comments! Follow me on Twitter for more tips and insights on coding and productivity. Follow me on Twitter

The Odin Project: My 111-Day Experience With It

ওডিন প্রজেক্ট: এটির সাথে আমার 111-দিনের অভিজ্ঞতা

Das Odin-Projekt: Meine 111-tägige Erfahrung damit

El Proyecto Odin: Mi experiencia de 111 días con él

Le projet Odin : mon expérience de 111 jours avec celui-ci

ओडिन परियोजना: इसके साथ मेरा 111-दिवसीय अनुभव

オーディン プロジェクト: 111 日間の体験

오딘 프로젝트: 나의 111일간의 경험

O Projeto Odin: Minha experiência de 111 dias com ele

Проект Odin: мой 111-дневный опыт работы с ним

Odin Projesi: 111 Günlük Deneyimim

Dự án Odin: Trải nghiệm 111 ngày của tôi với nó

奥丁计划：我的 111 天经历

Authors:
(1) Hamed Alimohammadzadeh, University of Southern California, Los Angeles, USA;
(2) Rohit Bernard, University of Southern California, Los Angeles, USA;
(3) Yang Chen, University of Southern California, Los Angeles, USA;
(4) Trung Phan, University of Southern California, Los Angeles, USA;
(5) Prashant Singh, University of Southern California, Los Angeles, USA;
(6) Shuqin Zhu, University of Southern California, Los Angeles, USA;
(7) Heather Culbertson, University of Southern California, Los Angeles, USA;
(8) Shahram Ghandeharizadeh, University of Southern California, Los Angeles, USA. Table of Links Abstract and Introduction Flying Light Specks User Interaction Multimedia Systems Challenges Related Work Conclusions and Current Efforts, Acknowledgments, and References 5 RELATED WORK A DV is in the same class of systems as fast 3D printing [21], Claytronics as physical artifacts using programmable matter consisting of catoms [28], Roboxels as cellular robots that dynamically configure themselves into the desired shape and size [42], BitDrones [29] and GridDrones [11] as interactive nano-drones, and Flying Light Specks (FLSs) as miniature drones with RGB light sources that fly as swarms to illuminate a virtual object [3, 25–27]. These studies use either analytical models, simulation studies, a lab setting, or a hybrid of these to demonstrate their design decisions and to quantify their tradeoffs. A DV is a much needed experimental testbed to implement and evaluate these techniques for multimedia applications using drones. Its modular design decisions and haptic user interactions raise many challenging multimedia systems research topics. If successful, it will contribute to the multimedia content creator’s experience in a rich and meaningful fashion while preserving their safety. It enables them to develop safe applications with complex and intertwined instances of different kinds of information including 3D acoustics, illuminations, and haptic interactions. The concept of 3D displays using FLSs is introduced in [25]. These FLSs require sensors to measure small distances in order to create 3D formations for visual and haptic displays. An evaluation of three different types of sensors to measure 1 centimeter distances is presented in [52]. User safety during interaction with a room size display with materialized objects is also important to creating effective interactions; we discuss potential safety considerations in [27]. An architecture of an FLS system is presented in [26], which introduces key components necessary to realize a successful implementation, including Motill to compute the flight path of FLSs to render a motion illumination, STAG as a battery charging technique, and use of standby FLSs to tolerate FLS failures [26]. This architecture may use the data model of [3] to store the flight paths of an algorithm such as Motill in a file. A sequence that is displayed repeatedly may read the file instead of executing the resource intensive Motill algorithm [3]. A DV is a small-scale experimental testbed for FLSs to realize the vision of a Holodeck in science fiction shows and as described in [25]. It builds on these prior studies with its novel use of disaggregated FLSs and a modular framework that emphasizes user safety. Additionally, a DV uses a swarm of FLSs to amplify force for haptic interactions, creating the important and novel challenges of localization and acoustics. This paper is available on arxiv under CC 4.0 license. Authors: (1) Hamed Alimohammadzadeh, University of Southern California, Los Angeles, USA; (2) Rohit Bernard, University of Southern California, Los Angeles, USA; (3) Yang Chen, University of Southern California, Los Angeles, USA; (4) Trung Phan, University of Southern California, Los Angeles, USA; (5) Prashant Singh, University of Southern California, Los Angeles, USA; (6) Shuqin Zhu, University of Southern California, Los Angeles, USA; (7) Heather Culbertson, University of Southern California, Los Angeles, USA; (8) Shahram Ghandeharizadeh, University of Southern California, Los Angeles, USA. Authors: Authors: (1) Hamed Alimohammadzadeh, University of Southern California, Los Angeles, USA; (2) Rohit Bernard, University of Southern California, Los Angeles, USA; (3) Yang Chen, University of Southern California, Los Angeles, USA; (4) Trung Phan, University of Southern California, Los Angeles, USA; (5) Prashant Singh, University of Southern California, Los Angeles, USA; (6) Shuqin Zhu, University of Southern California, Los Angeles, USA; (7) Heather Culbertson, University of Southern California, Los Angeles, USA; (8) Shahram Ghandeharizadeh, University of Southern California, Los Angeles, USA. Table of Links Abstract and Introduction Abstract and Introduction Flying Light Specks Flying Light Specks User Interaction User Interaction Multimedia Systems Challenges Multimedia Systems Challenges Related Work Related Work Conclusions and Current Efforts, Acknowledgments, and References Conclusions and Current Efforts, Acknowledgments, and References 5 RELATED WORK A DV is in the same class of systems as fast 3D printing [21], Claytronics as physical artifacts using programmable matter consisting of catoms [28], Roboxels as cellular robots that dynamically configure themselves into the desired shape and size [42], BitDrones [29] and GridDrones [11] as interactive nano-drones, and Flying Light Specks (FLSs) as miniature drones with RGB light sources that fly as swarms to illuminate a virtual object [3, 25–27]. These studies use either analytical models, simulation studies, a lab setting, or a hybrid of these to demonstrate their design decisions and to quantify their tradeoffs. A DV is a much needed experimental testbed to implement and evaluate these techniques for multimedia applications using drones. Its modular design decisions and haptic user interactions raise many challenging multimedia systems research topics. If successful, it will contribute to the multimedia content creator’s experience in a rich and meaningful fashion while preserving their safety. It enables them to develop safe applications with complex and intertwined instances of different kinds of information including 3D acoustics, illuminations, and haptic interactions. The concept of 3D displays using FLSs is introduced in [25]. These FLSs require sensors to measure small distances in order to create 3D formations for visual and haptic displays. An evaluation of three different types of sensors to measure 1 centimeter distances is presented in [52]. User safety during interaction with a room size display with materialized objects is also important to creating effective interactions; we discuss potential safety considerations in [27]. An architecture of an FLS system is presented in [26], which introduces key components necessary to realize a successful implementation, including Motill to compute the flight path of FLSs to render a motion illumination, STAG as a battery charging technique, and use of standby FLSs to tolerate FLS failures [26]. This architecture may use the data model of [3] to store the flight paths of an algorithm such as Motill in a file. A sequence that is displayed repeatedly may read the file instead of executing the resource intensive Motill algorithm [3]. A DV is a small-scale experimental testbed for FLSs to realize the vision of a Holodeck in science fiction shows and as described in [25]. It builds on these prior studies with its novel use of disaggregated FLSs and a modular framework that emphasizes user safety. Additionally, a DV uses a swarm of FLSs to amplify force for haptic interactions, creating the important and novel challenges of localization and acoustics. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

Stories, News & Mentions about EXPERIFY

Experify’s Stories on HackerNoon