EBIS

ebis

Enabling Acoustic Audience Feedback in Large Virtual Events: Conclusions & References

SearchPedia: A List of 250+ Search Engines

SearchPedia: una lista de más de 250 motores de búsqueda

Authors:
(1) Luyang Zhu, University of Washington and Google Research, and work done while the author was an intern at Google;
(2) Dawei Yang, Google Research;
(3) Tyler Zhu, Google Research;
(4) Fitsum Reda, Google Research;
(5) William Chan, Google Research;
(6) Chitwan Saharia, Google Research;
(7) Mohammad Norouzi, Google Research;
(8) Ira Kemelmacher-Shlizerman, University of Washington and Google Research. Table of Links Abstract and 1. Introduction 2. Related Work 3. Method 3.1. Cascaded Diffusion Models for Try-On 3.2. Parallel-UNet 4. Experiments 5. Summary and Future Work and References Appendix A. Implementation Details B. Additional Results 2. Related Work Image-Based Virtual Try-On. Given a pair of images (target person, source garment), image-based virtual try-on methods generate the look of the target person wearing the source garment. Most of these methods [2, 6, 7, 10, 14, 15, 20,25,27,32,43,46–49] decompose the try-on task into two stages: a warping model and a blending model. The seminal work VITON [14] proposes a coarse-to-fine pipeline guided by the thin-plate-spline (TPS) warping of source garments. ClothFlow [13] directly estimates flow fields with a neural network instead of TPS for better garment warping. VITON-HD [6] introduces alignment-aware generator to increase the try-on resolution from 256×192 to 1024×768. HR-VITON [25] further improves VITON-HD by predicting segmentation and flow simultaneously. SDAFN [2] predicts multiple flow fields for both the garment and the person, and combines warped features through deformable attention [50] to improve quality. Despite great progress, these methods still suffer from misalignment brought by explicit flow estimation and warping. TryOnGAN [26] tackles this issue by training a poseconditioned StyleGAN2 [23] on unpaired fashion images and running optimization in the latent space to achieve tryon. By optimizing the latent space, however, it loses garment details that are less represented by the latent space. This becomes evident when garments have a pattern or details like pockets, or special sleeves. We propose a novel architecture which performs implicit warping (without computing flow) and blending in a single network pass. Experiments show that our method can preserve details of the garment even under heavy occlusions and various body poses and shapes. Diffusion Models. Diffusion models [17, 39, 41] have recently emerged as the most powerful family of generative models. Unlike GANs [5, 12], diffusion models have better training stability and mode coverage. They have achieved state-of-the-art results on various image generation tasks, such as super-resolution [38], colorization [36], novel-view synthesis [44] and text-to-image generation [30, 33, 35, 37]. Although being successful, state-of-the-art diffusion models utilize a traditional UNet architecture [17, 34] with channel-wise concatenation [36,38] for image conditioning. The channel-wise concatenation works well for image-toimage translation problems where input and output pixels are perfectly aligned (e.g., super-resolution, inpainting and colorization). However, it is not directly applicable to our task as try-on involves highly non-linear transformations like garment warping. To solve this challenge, we propose Parallel-UNet architecture tailored to try-on, where the garment is warped implicitly via cross attentions. This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. Authors: (1) Luyang Zhu, University of Washington and Google Research, and work done while the author was an intern at Google; (2) Dawei Yang, Google Research; (3) Tyler Zhu, Google Research; (4) Fitsum Reda, Google Research; (5) William Chan, Google Research; (6) Chitwan Saharia, Google Research; (7) Mohammad Norouzi, Google Research; (8) Ira Kemelmacher-Shlizerman, University of Washington and Google Research. Authors: Authors: (1) Luyang Zhu, University of Washington and Google Research, and work done while the author was an intern at Google; (2) Dawei Yang, Google Research; (3) Tyler Zhu, Google Research; (4) Fitsum Reda, Google Research; (5) William Chan, Google Research; (6) Chitwan Saharia, Google Research; (7) Mohammad Norouzi, Google Research; (8) Ira Kemelmacher-Shlizerman, University of Washington and Google Research. Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2. Related Work 2. Related Work 3. Method 3. Method 3.1. Cascaded Diffusion Models for Try-On 3.1. Cascaded Diffusion Models for Try-On 3.2. Parallel-UNet 3.2. Parallel-UNet 4. Experiments 4. Experiments 5. Summary and Future Work and References 5. Summary and Future Work and References Appendix Appendix A. Implementation Details A. Implementation Details B. Additional Results B. Additional Results 2. Related Work Image-Based Virtual Try-On. Given a pair of images (target person, source garment), image-based virtual try-on methods generate the look of the target person wearing the source garment. Most of these methods [2, 6, 7, 10, 14, 15, 20,25,27,32,43,46–49] decompose the try-on task into two stages: a warping model and a blending model. The seminal work VITON [14] proposes a coarse-to-fine pipeline guided by the thin-plate-spline (TPS) warping of source garments. ClothFlow [13] directly estimates flow fields with a neural network instead of TPS for better garment warping. VITON-HD [6] introduces alignment-aware generator to increase the try-on resolution from 256×192 to 1024×768. HR-VITON [25] further improves VITON-HD by predicting segmentation and flow simultaneously. SDAFN [2] predicts multiple flow fields for both the garment and the person, and combines warped features through deformable attention [50] to improve quality. Despite great progress, these methods still suffer from misalignment brought by explicit flow estimation and warping. TryOnGAN [26] tackles this issue by training a poseconditioned StyleGAN2 [23] on unpaired fashion images and running optimization in the latent space to achieve tryon. By optimizing the latent space, however, it loses garment details that are less represented by the latent space. This becomes evident when garments have a pattern or details like pockets, or special sleeves. We propose a novel architecture which performs implicit warping (without computing flow) and blending in a single network pass. Experiments show that our method can preserve details of the garment even under heavy occlusions and various body poses and shapes. Diffusion Models. Diffusion models [17, 39, 41] have recently emerged as the most powerful family of generative models. Unlike GANs [5, 12], diffusion models have better training stability and mode coverage. They have achieved state-of-the-art results on various image generation tasks, such as super-resolution [38], colorization [36], novel-view synthesis [44] and text-to-image generation [30, 33, 35, 37]. Although being successful, state-of-the-art diffusion models utilize a traditional UNet architecture [17, 34] with channel-wise concatenation [36,38] for image conditioning. The channel-wise concatenation works well for image-toimage translation problems where input and output pixels are perfectly aligned (e.g., super-resolution, inpainting and colorization). However, it is not directly applicable to our task as try-on involves highly non-linear transformations like garment warping. To solve this challenge, we propose Parallel-UNet architecture tailored to try-on, where the garment is warped implicitly via cross attentions. Diffusion Models. 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

How TryOnDiffusion Innovates on Existing Virtual Try-On Frameworks

**PALO ALTO, CA, USA, October 1st, 2024/GamingWire/--**Via a partnership with Sui original contributor Mysten Labs, Gamisodes releases remastered mobile-playable episodes of Inspector Gadget with digital rewards. Mysten Labs, a web3 infrastructure company and the original contributor to the Sui blockchain, today announced a partnership with Gamisodes, a streaming platform that transforms classic TV episodes into playable games. As of today, Gamisodes has launched the first remastered playable episode of the beloved animated classic, Inspector Gadget, which incorporates digital assets powered by Sui’s blockchain technology. Gamisodes has remastered episodes from the 1983 television series, Inspector Gadget, and transformed them into playable mobile games, allowing players to own unique pieces of pop culture history. Through the use of Sui’s innovative features, players can engage in immersive gameplay and will be provided with seamless digital ownership by unlocking exclusive in-game assets including avatars and collectibles. “Interactive entertainment is a perfect example of the innovative technologies we want to empower through the decentralized internet,” said Adeniyi Abidoun, CPO and Co-Founder of Mysten Labs. “Gamisodes is bringing the proof of digital ownership concept on-chain, which is a pivotal step in empowering creators to continue building more engaging decentralized experiences.” The first three episodes of Inspector Gadget will be available to play for free through the Gamisodes application available on iOS and Android in all territories worldwide. Following episodes will be released over the next several months, during which players will be rewarded with exclusive virtual store items, virtual and live events, giveaways, and surprise guests. The first episode will feature single player gameplay, with further gameplay structure to be revealed soon. “Audiences today crave deeper, more engaging forms of entertainment, and we’re excited to deliver exactly that. Gamisodes blends storytelling with mobile-first, interactive gameplay, creating an experience that’s not just watched, but played and owned,” said Davis Brimer, Founder & CEO of Gamisodes. “Thanks to Sui’s innovative blockchain technology, the ownership aspect is seamless and secure.” For additional information on pricing and upcoming releases, please visit https://gamisodes.com. About Gamisodes For digitally native audiences that are left wanting more interactive and engaging entertainment, Gamisodes is a first-of-its-kind interactive entertainment platform that streams TV entertainment and mobile gameplay within a single offering. Unlike services that keep TV and interactive separate, as two distinct experiences, Gamisodes brings them together – delivering the next stage in the evolution of streaming: TV Episode Gameplay. Gamisodes’ flagship interactive show, the OG 1980s classic Inspector Gadget, is out now on the free Gamisodes app – available on iOS and Android. To learn more: https://gamisodes.com About Sui Sui is a first-of-its-kind Layer 1 blockchain and smart contract platform designed from the bottom up to make digital asset ownership fast, private, secure, and accessible to everyone. Its object-centric model, based on the Move programming language, enables parallel execution, sub-second finality, and rich on-chain assets. With horizontally scalable processing and storage, Sui supports a wide range of applications with unrivaled speed at low cost. Sui is a step-function advancement in blockchain and a platform on which creators and developers can build amazing, user-friendly experiences. To learn more: https://sui.io About Mysten Labs Mysten Labs is a team of leading distributed systems, programming languages, and cryptography experts whose founders were senior executives and lead architects of pioneering blockchain projects. The mission of Mysten Labs is to create foundational infrastructure for web3. To learn more: https://mystenlabs.com Contact Lexi Wangler Mysten Labs lexi.wangler@mystenlabs.com This story was distributed as a release by Chainwire under HackerNoon’s Business Blogging Program. Learn more about the program here **PALO ALTO, CA, USA, October 1st, 2024/GamingWire/--**Via a partnership with Sui original contributor Mysten Labs, Gamisodes releases remastered mobile-playable episodes of Inspector Gadget with digital rewards. Mysten Labs , a web3 infrastructure company and the original contributor to the Sui blockchain , today announced a partnership with Gamisodes , a streaming platform that transforms classic TV episodes into playable games. As of today, Gamisodes has launched the first remastered playable episode of the beloved animated classic, Inspector Gadget, which incorporates digital assets powered by Sui’s blockchain technology. Mysten Labs Mysten Labs Sui blockchain Sui blockchain Gamisodes Gamisodes Gamisodes has remastered episodes from the 1983 television series, Inspector Gadget, and transformed them into playable mobile games, allowing players to own unique pieces of pop culture history. Through the use of Sui’s innovative features, players can engage in immersive gameplay and will be provided with seamless digital ownership by unlocking exclusive in-game assets including avatars and collectibles. “Interactive entertainment is a perfect example of the innovative technologies we want to empower through the decentralized internet,” said Adeniyi Abidoun, CPO and Co-Founder of Mysten Labs. “Gamisodes is bringing the proof of digital ownership concept on-chain, which is a pivotal step in empowering creators to continue building more engaging decentralized experiences.” “Interactive entertainment is a perfect example of the innovative technologies we want to empower through the decentralized internet,” said Adeniyi Abidoun, CPO and Co-Founder of Mysten Labs. “Gamisodes is bringing the proof of digital ownership concept on-chain, which is a pivotal step in empowering creators to continue building more engaging decentralized experiences.” The first three episodes of Inspector Gadget will be available to play for free through the Gamisodes application available on iOS and Android in all territories worldwide. Following episodes will be released over the next several months, during which players will be rewarded with exclusive virtual store items, virtual and live events, giveaways, and surprise guests. The first episode will feature single player gameplay, with further gameplay structure to be revealed soon. “Audiences today crave deeper, more engaging forms of entertainment, and we’re excited to deliver exactly that. Gamisodes blends storytelling with mobile-first, interactive gameplay, creating an experience that’s not just watched, but played and owned,” said Davis Brimer, Founder & CEO of Gamisodes. “Thanks to Sui’s innovative blockchain technology, the ownership aspect is seamless and secure.” For additional information on pricing and upcoming releases, please visit https://gamisodes.com . https://gamisodes.com https://gamisodes.com About Gamisodes For digitally native audiences that are left wanting more interactive and engaging entertainment, Gamisodes is a first-of-its-kind interactive entertainment platform that streams TV entertainment and mobile gameplay within a single offering. Unlike services that keep TV and interactive separate, as two distinct experiences, Gamisodes brings them together – delivering the next stage in the evolution of streaming: TV Episode Gameplay. Gamisodes’ flagship interactive show, the OG 1980s classic Inspector Gadget, is out now on the free Gamisodes app – available on iOS and Android. To learn more: https://gamisodes.com https://gamisodes.com https://gamisodes.com About Sui Sui is a first-of-its-kind Layer 1 blockchain and smart contract platform designed from the bottom up to make digital asset ownership fast, private, secure, and accessible to everyone. Its object-centric model, based on the Move programming language, enables parallel execution, sub-second finality, and rich on-chain assets. With horizontally scalable processing and storage, Sui supports a wide range of applications with unrivaled speed at low cost. Sui is a step-function advancement in blockchain and a platform on which creators and developers can build amazing, user-friendly experiences. To learn more: https://sui.io https://sui.io https://sui.io About Mysten Labs Mysten Labs is a team of leading distributed systems, programming languages, and cryptography experts whose founders were senior executives and lead architects of pioneering blockchain projects. The mission of Mysten Labs is to create foundational infrastructure for web3. To learn more: https://mystenlabs.com https://mystenlabs.com https://mystenlabs.com Contact Lexi Wangler Mysten Labs lexi.wangler@mystenlabs.com This story was distributed as a release by Chainwire under HackerNoon’s Business Blogging Program. Learn more about the program here This story was distributed as a release by Chainwire under HackerNoon’s Business Blogging Program. Learn more about the program here here here

Gamisodes Launches Playable TV Episodes, Digital Ownership Made Possible By Sui

Gamisodes stel speelbare TV-episodes bekend, digitale eienaarskap wat deur Sui moontlik gemaak is

Gamisodes Oynana bilən TV Epizodlarını təqdim edir, Rəqəmsal Sahiblik Sui tərəfindən Mümkündür

Gamisodes lanza episodios de televisión jugables, propiedad digital posible gracias a Sui

A Gamisodes lejátszható TV-epizódokat indít, a digitális tulajdonjogot a Sui tette lehetővé

Gamisodes-ը թողարկում է նվագարկվող հեռուստատեսային դրվագներ, թվային սեփականություն, որը հնարավոր է դարձել Sui-ի կողմից

Gamisodesがプレイ可能なTVエピソードを開始、Suiによるデジタル所有権が可能に

Gamisodes ойнотула турган ТВ эпизоддорун чыгарат, санариптик ээлик Sui тарабынан мүмкүн болгон

„Gamisodes“ pristato leidžiamus TV epizodus, skaitmeninę nuosavybę suteikė Sui

Gamisodes izlaiž atskaņojamas TV sērijas, un digitālās īpašumtiesības nodrošina Sui

Gamisodes Melancarkan Episod TV Boleh Main, Pemilikan Digital Dimungkinkan Oleh Sui

Gamisodes Yazindua Vipindi vya Televisheni Vinavyoweza Kuchezwa, Umiliki Dijitali Wawezeshwa na Sui

Gamisodes запускає відтворювані телевізійні серії, цифрове володіння стало можливим завдяки Sui

Effective Security Operations Center (SOC) management is more complex than ever. Analysts are plagued by alert fatigue, threats are growing more frequent and sophisticated, and enterprise environments are becoming increasingly complex. However, these challenges are not insurmountable: intelligent use of the right security tools can significantly reduce the burden on SOC teams. Security Orchestration, Automation, and Response (SOAR) is one such tool. What is SOAR? At root, SOAR solutions collect data and respond to security events without human intervention. As the name suggests, they have three key capabilities: orchestration, automation, and response. Before we cover how best to integrate SOAR into your existing security program, it’s crucial to understand these terms. Orchestration Security orchestration involves coordinating and integrating multiple security tools and technologies. Just as a musical orchestra must work harmoniously, so does a security stack. SOAR solutions pull data from an organization’s many security tools – including everything from firewalls to threat intelligence platforms – and display this information on a centralized interface, ensuring security teams have a comprehensive view of their security environment and, ultimately, better protect their organization. Automation It’s a hot topic at the moment—and one you are probably already familiar with. Fundamentally, security automation refers to the pre-programmed execution of security actions. SOAR solutions create and execute processes to replace manual actions based on the data collected during security orchestration. For example, the SOAR tool takes over tasks usually performed by security analysts—such as vulnerability scanning or log analysis. Playbooks are critical to the automation process. They contain predefined workflows that the SOAR solution executes in response to specific incidents. For example, suppose the solution determines during the orchestration phase that someone has gained unauthorized access to the network. In that case, it will trigger response actions that isolate, quarantine, and elevate the threat. Response SOAR’s response capabilities are an amalgamation of its orchestration and automation capabilities. Orchestration gives security teams a single view of the security environment, while automation executes immediate response actions. The solution provides data for further investigation in the aftermath of an incident. How to Integrate Your SOAR Solution Now that we understand how a SOAR solution works, we can explore how best to integrate it with your existing security tools. However, it’s crucial to remember that SOAR integration is a complicated process that will take months to complete. This blog is not a comprehensive guide to SOAR integration – think of it as a jumping-off point to inform the remainder of your integration journey. Assess and Map Your Current Security Infrastructure First, you must inventory all your existing security tools – and that means all of them. Your inventory should include information on your tools’ functionalities, data flows, and integration capabilities. It’s also worth prioritizing tools for integration, focusing on those most critical for incident detection and response. Select the Right SOAR Platform Once you have inventoried your tools, you can start thinking about what SOAR platform is best for you. You must ensure the SOAR you choose supports integrations with your existing tools, either natively or via APIs, allows you to customize workflows, playbooks, and integrations, and can scale with your business. Talk with multiple SOAR providers and ask these questions during the buying process. Plan and Prioritize Integration You will have done some of this in the inventory stage, but planning and prioritizing your integration process is now essential. Identify the use cases – such as incident response and threat hunting – where SOAR will help most. Similarly, you must set clear objectives for your SOAR integration – such as reducing response times – and prioritize the integrations that will best help you achieve them. Implement and Configure Integrations We’re finally past the planning stage and into the actual integration. During this process, you must use APIs or custom connectors to link your security tools with the SOAR platform – this will ensure proper security orchestration. Then, you must develop playbooks that define automated response actions for specific incidents. Most importantly, test your integrations before implementation to ensure everything functions as it should. Monitor, Optimize, and Maintain At this stage, your SOAR will be set up and ready to go. But your work isn’t over: continuous improvement is crucial to the success of a SOAR solution. You must track metrics like mean time to detect (MTTD) and mean time to respond (MTTR) to evaluate the effectiveness of your SOAR integration. You must also regularly update and refine playbooks based on new threats, lessons learned from incidents, and changes in your security environment to ensure your solution works as it should. Conclusion Security Orchestration, Automation, and Response are vital parts of any effective SOC. They significantly reduce alert fatigue among SOC analysts, streamline work processes, offer a comprehensive view of security environments, and ultimately improve security. However, integrating SOAR with existing security systems is no easy task; it requires a thoughtful approach. Use the best practices above to inform your integration process and work with your SOAR provider for successful integration. Effective Security Operations Center ( SOC ) management is more complex than ever. Analysts are plagued by alert fatigue, threats are growing more frequent and sophisticated, and enterprise environments are becoming increasingly complex. However, these challenges are not insurmountable: intelligent use of the right security tools can significantly reduce the burden on SOC teams. Security Orchestration, Automation, and Response (SOAR) is one such tool. SOC SOC What is SOAR? At root, SOAR solutions collect data and respond to security events without human intervention. As the name suggests, they have three key capabilities: orchestration, automation, and response. Before we cover how best to integrate SOAR into your existing security program, it’s crucial to understand these terms. Orchestration Security orchestration involves coordinating and integrating multiple security tools and technologies. Just as a musical orchestra must work harmoniously, so does a security stack. SOAR solutions pull data from an organization’s many security tools – including everything from firewalls to threat intelligence platforms – and display this information on a centralized interface, ensuring security teams have a comprehensive view of their security environment and, ultimately, better protect their organization. Automation It’s a hot topic at the moment—and one you are probably already familiar with. Fundamentally, security automation refers to the pre-programmed execution of security actions. SOAR solutions create and execute processes to replace manual actions based on the data collected during security orchestration. For example, the SOAR tool takes over tasks usually performed by security analysts—such as vulnerability scanning or log analysis. vulnerability scanning vulnerability scanning Playbooks are critical to the automation process. They contain predefined workflows that the SOAR solution executes in response to specific incidents. For example, suppose the solution determines during the orchestration phase that someone has gained unauthorized access to the network. In that case, it will trigger response actions that isolate, quarantine, and elevate the threat. Response SOAR’s response capabilities are an amalgamation of its orchestration and automation capabilities. Orchestration gives security teams a single view of the security environment, while automation executes immediate response actions. The solution provides data for further investigation in the aftermath of an incident. How to Integrate Your SOAR Solution Now that we understand how a SOAR solution works, we can explore how best to integrate it with your existing security tools. However, it’s crucial to remember that SOAR integration is a complicated process that will take months to complete. This blog is not a comprehensive guide to SOAR integration – think of it as a jumping-off point to inform the remainder of your integration journey. Assess and Map Your Current Security Infrastructure First, you must inventory all your existing security tools – and that means all of them. Your inventory should include information on your tools’ functionalities, data flows, and integration capabilities. It’s also worth prioritizing tools for integration, focusing on those most critical for incident detection and response. Select the Right SOAR Platform Once you have inventoried your tools, you can start thinking about what SOAR platform is best for you. You must ensure the SOAR you choose supports integrations with your existing tools, either natively or via APIs, allows you to customize workflows, playbooks, and integrations, and can scale with your business. Talk with multiple SOAR providers and ask these questions during the buying process. Plan and Prioritize Integration You will have done some of this in the inventory stage, but planning and prioritizing your integration process is now essential. Identify the use cases – such as incident response and threat hunting – where SOAR will help most. Similarly, you must set clear objectives for your SOAR integration – such as reducing response times – and prioritize the integrations that will best help you achieve them. Implement and Configure Integrations We’re finally past the planning stage and into the actual integration. During this process, you must use APIs or custom connectors to link your security tools with the SOAR platform – this will ensure proper security orchestration. Then, you must develop playbooks that define automated response actions for specific incidents. Most importantly, test your integrations before implementation to ensure everything functions as it should. Monitor, Optimize, and Maintain At this stage, your SOAR will be set up and ready to go. But your work isn’t over: continuous improvement is crucial to the success of a SOAR solution. You must track metrics like mean time to detect (MTTD) and mean time to respond (MTTR) to evaluate the effectiveness of your SOAR integration. You must also regularly update and refine playbooks based on new threats, lessons learned from incidents, and changes in your security environment to ensure your solution works as it should. mean time to respond mean time to respond Conclusion Security Orchestration, Automation, and Response are vital parts of any effective SOC. They significantly reduce alert fatigue among SOC analysts, streamline work processes, offer a comprehensive view of security environments, and ultimately improve security. However, integrating SOAR with existing security systems is no easy task; it requires a thoughtful approach. Use the best practices above to inform your integration process and work with your SOAR provider for successful integration. alert fatigue alert fatigue

Integrating SOAR with Existing Security Systems

Authors:
(1) Suriya Gunasekar, Microsoft Research;
(2) Yi Zhang, Microsoft Research;
(3) Jyoti Aneja, Microsoft Research;
(4) Caio C´esar Teodoro Mendes, Microsoft Research;
(5) Allie Del Giorno, Microsoft Research;
(6) Sivakanth Gopi, Microsoft Research;
(7) Mojan Javaheripi, Microsoft Research;
(8) Piero Kauffmann, Microsoft Research;
(9) Gustavo de Rosa, Microsoft Research;
(10) Olli Saarikivi, Microsoft Research;
(11) Adil Salim, Microsoft Research;
(12) Shital Shah, Microsoft Research;
(13) Harkirat Singh Behl, Microsoft Research;
(14) Xin Wang, Microsoft Research;
(15) S´ebastien Bubeck, Microsoft Research;
(16) Ronen Eldan, Microsoft Research;
(17) Adam Tauman Kalai, Microsoft Research;
(18) Yin Tat Lee, Microsoft Research;
(19) Yuanzhi Li, Microsoft Research. Table of Links Abstract and 1. Introduction
2 Training details and the importance of high-quality data
2.1 Filtering of existing code datasets using a transformer-based classifier
2.2 Creation of synthetic textbook-quality datasets
2.3 Model architecture and training
3 Spikes of model capability after finetuning on CodeExercises, 3.1 Finetuning improves the model’s understanding, and 3.2 Finetuning improves the model’s ability to use external libraries
4 Evaluation on unconventional problems with LLM grading
5 Data pruning for unbiased performance evaluation
5.1 N-gram overlap and 5.2 Embedding and syntax-based similarity analysis
6 Conclusion and References
A Additional examples for Section 3
B Limitation of phi-1
C Examples for Section 5 2.1 Filtering of existing code datasets using a transformer-based classifier We begin with publicly available Python code datasets: we use the Python subset of the deduplicated version of The Stack and the StackOverflow, which together contain over 35 million files/samples, totalling over 35B tokens. We annotate the quality of a small subset of these files (about 100k samples) using GPT-4: given a code snippet, the model is prompted to “determine its educational value for a student whose goal is to learn basic coding concepts”. We then use this annotated dataset to train a random forest classifier that predicts the quality of a file/sample using its output embedding from a pretrained codegen model as features. We note that unlike GPT-3.5, which we use extensively to generate synthetic content (discussed below), we use GPT-4 minimally only for annotations on the quality of a small subset of The Stack and StackOverflow samples. We thus view our usage of GPT-4 as merely a way to avoid tedious human-annotation efforts [DLT+ 23]. Our filtering methodology boosts our model performance significantly even without the synthetic datasets discussed below: for 350M parameter models trained on unfiltered Stack (deduplicated python) and StackOverflow, the HumanEval performance saturates at 12.19% even after training for 96k steps (∼ 200B tokens), while training on the filtered subset achieves 17.68% on HumanEval after 36k steps. We further improve this to 20.12% (reported in Figure 2.1) by training on a combination of the filtered dataset and the synthetic textbooks dataset discussed below. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Suriya Gunasekar, Microsoft Research; (2) Yi Zhang, Microsoft Research; (3) Jyoti Aneja, Microsoft Research; (4) Caio C´esar Teodoro Mendes, Microsoft Research; (5) Allie Del Giorno, Microsoft Research; (6) Sivakanth Gopi, Microsoft Research; (7) Mojan Javaheripi, Microsoft Research; (8) Piero Kauffmann, Microsoft Research; (9) Gustavo de Rosa, Microsoft Research; (10) Olli Saarikivi, Microsoft Research; (11) Adil Salim, Microsoft Research; (12) Shital Shah, Microsoft Research; (13) Harkirat Singh Behl, Microsoft Research; (14) Xin Wang, Microsoft Research; (15) S´ebastien Bubeck, Microsoft Research; (16) Ronen Eldan, Microsoft Research; (17) Adam Tauman Kalai, Microsoft Research; (18) Yin Tat Lee, Microsoft Research; (19) Yuanzhi Li, Microsoft Research. Authors: Authors: (1) Suriya Gunasekar, Microsoft Research; (2) Yi Zhang, Microsoft Research; (3) Jyoti Aneja, Microsoft Research; (4) Caio C´esar Teodoro Mendes, Microsoft Research; (5) Allie Del Giorno, Microsoft Research; (6) Sivakanth Gopi, Microsoft Research; (7) Mojan Javaheripi, Microsoft Research; (8) Piero Kauffmann, Microsoft Research; (9) Gustavo de Rosa, Microsoft Research; (10) Olli Saarikivi, Microsoft Research; (11) Adil Salim, Microsoft Research; (12) Shital Shah, Microsoft Research; (13) Harkirat Singh Behl, Microsoft Research; (14) Xin Wang, Microsoft Research; (15) S´ebastien Bubeck, Microsoft Research; (16) Ronen Eldan, Microsoft Research; (17) Adam Tauman Kalai, Microsoft Research; (18) Yin Tat Lee, Microsoft Research; (19) Yuanzhi Li, Microsoft Research. Table of Links Abstract and 1. Introduction 2 Training details and the importance of high-quality data 2.1 Filtering of existing code datasets using a transformer-based classifier 2.2 Creation of synthetic textbook-quality datasets 2.3 Model architecture and training 3 Spikes of model capability after finetuning on CodeExercises, 3.1 Finetuning improves the model’s understanding, and 3.2 Finetuning improves the model’s ability to use external libraries 4 Evaluation on unconventional problems with LLM grading 5 Data pruning for unbiased performance evaluation 5.1 N-gram overlap and 5.2 Embedding and syntax-based similarity analysis 6 Conclusion and References A Additional examples for Section 3 B Limitation of phi-1 C Examples for Section 5 Abstract and 1. Introduction Abstract and 1. Introduction 2 Training details and the importance of high-quality data 2 Training details and the importance of high-quality data 2.1 Filtering of existing code datasets using a transformer-based classifier 2.1 Filtering of existing code datasets using a transformer-based classifier 2.2 Creation of synthetic textbook-quality datasets 2.2 Creation of synthetic textbook-quality datasets 2.3 Model architecture and training 2.3 Model architecture and training 3 Spikes of model capability after finetuning on CodeExercises, 3.1 Finetuning improves the model’s understanding, and 3.2 Finetuning improves the model’s ability to use external libraries 3 Spikes of model capability after finetuning on CodeExercises, 3.1 Finetuning improves the model’s understanding, and 3.2 Finetuning improves the model’s ability to use external libraries 4 Evaluation on unconventional problems with LLM grading 4 Evaluation on unconventional problems with LLM grading 5 Data pruning for unbiased performance evaluation 5 Data pruning for unbiased performance evaluation 5.1 N-gram overlap and 5.2 Embedding and syntax-based similarity analysis 5.1 N-gram overlap and 5.2 Embedding and syntax-based similarity analysis 6 Conclusion and References 6 Conclusion and References A Additional examples for Section 3 A Additional examples for Section 3 B Limitation of phi-1 B Limitation of phi-1 C Examples for Section 5 C Examples for Section 5 2.1 Filtering of existing code datasets using a transformer-based classifier We begin with publicly available Python code datasets: we use the Python subset of the deduplicated version of The Stack and the StackOverflow, which together contain over 35 million files/samples, totalling over 35B tokens. We annotate the quality of a small subset of these files (about 100k samples) using GPT-4: given a code snippet, the model is prompted to “determine its educational value for a student whose goal is to learn basic coding concepts”. We then use this annotated dataset to train a random forest classifier that predicts the quality of a file/sample using its output embedding from a pretrained codegen model as features. We note that unlike GPT-3.5, which we use extensively to generate synthetic content (discussed below), we use GPT-4 minimally only for annotations on the quality of a small subset of The Stack and StackOverflow samples. We thus view our usage of GPT-4 as merely a way to avoid tedious human-annotation efforts [DLT+ 23]. Our filtering methodology boosts our model performance significantly even without the synthetic datasets discussed below: for 350M parameter models trained on unfiltered Stack (deduplicated python) and StackOverflow, the HumanEval performance saturates at 12.19% even after training for 96k steps (∼ 200B tokens), while training on the filtered subset achieves 17.68% on HumanEval after 36k steps. We further improve this to 20.12% (reported in Figure 2.1) by training on a combination of the filtered dataset and the synthetic textbooks dataset discussed below. 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

Textbooks are All You Need: Filtering of Existing Code Datasets Using a Transformer-based Classifier

Theoretical Physics: Gaps in Reality, Existence,  Panpsychism, and Biology

Questions that may be essential to AI regulation for now may include current and potential misuses, sources of those misuses, and why they are technically permissible. Broad AI regulations like those for energy, pharmaceuticals, and other sectors may apply in some form, but ultimately, AI regulation that may work would likely be technical. Physical laws can regulate digital products but would find limitations. There are several AI tools that misuses are possible from that may slip regulations. There are ways that some frontier AI models are self-regulating that may leave little room for adjustments from authorities. Even in the event of presenting models for safety evaluation before public release, frontier AI companies may have done some work, covering what they expect would be sought. How can AI regulation technically fight misuse? This includes the possibility to explore real-time multimodal misuses, across AI models. How is it possible to track and monitor some outputs of all AI tools that are available in a jurisdiction? California Senate has SB 1047 to regulate AI about "Giving companies more flexibility in how they meet their responsible development obligations while still holding them liable if irresponsible behavior leads to catastrophic harm. Allowing a state agency to change the compute threshold for a model to be covered under the bill starting in 2027, rather than keeping it fixed in statute at 10^26 flops. The requirement that the model cost at least $100 million to train cannot be changed. Requiring companies get third-party safety audits by 2028. Strengthen whistleblower protections. Requiring developers of such large “frontier” AI models to take basic precautions, such as pre-deployment safety testing, red-teaming, cybersecurity, safeguards to prevent the misuse of dangerous capabilities, and post-deployment monitoring. " The European Commission in AI Act enters into force, states that "On 1 August 2024, the European Artificial Intelligence Act (AI Act) enters into force. The Act aims to foster responsible artificial intelligence development and deployment in the EU. Proposed by the Commission in April 2021 and agreed by the European Parliament and the Council in December 2023, the AI Act addresses potential risks to citizens’ health, safety, and fundamental rights. It provides developers and deployers with clear requirements and obligations regarding specific uses of AI while reducing administrative and financial burdens for businesses. The AI Act introduces a uniform framework across all EU countries, based on a forward-looking definition of AI and a risk-based approach:" The problem with AI regulation is not that laws are not possible around it or that industry standards cannot be set. It is just that the ability to catch misuses on the go is better than regulation at the gate. Simply, it is vital to catch misuses, which in this case goes beyond the aerial view of frontier companies or start-ups but to tools, across usage sources, consistently. This will be like fighting digital piracy at any source it is available within a jurisdiction, not just on major platforms, or pointing to physical laws when digital evasion is knavish. There are research options like penalty-tuning and web scraping of available AI tools. Penalty-tuning could be a way to penalize LLMs, by compute, usage, or language, in a way that the model can know, if it outputs misinformation or deepfakes. Web scraping could be possible for AI tools that are available in search results so that some of the keywords around misuses can be tracked to expect them at their likely destinations, to prevent the harm they might cause. AI regulation is a problem of research first, then of laws. Exploring how to advance, technically, against misuses could guide what directions laws may take. There is a recent preprint on arXiv, Are Emergent Abilities in Large Language Models just In-Context Learning?, where the authors wrote, "We present a novel theory that explains emergent abilities, taking into account their potential confounding factors, and rigorously substantiate this theory through over 1000 experiments. Our findings suggest that purported emergent abilities are not truly emergent, but result from a combination of in-context learning, model memory, and linguistic knowledge. Our work is a foundational step in explaining language model performance, providing a template for their efficient use and clarifying the paradox of their ability to excel in some instances while faltering in others. Thus, we demonstrate that their capabilities should not be overestimated." AI is simply not an existential risk because of prediction-based LLMs. In the human mind, what is called prediction is a type of distribution or relay. LLMs have a substantial part of that, but they do not have what follows after [the incoming half of electrical signals, conceptually]. As AI gets better, with non-concept features and distributions similar to how relays within human memory result in human intelligence, they may pose higher levels of risks. Questions that may be essential to AI regulation for now may include current and potential misuses, sources of those misuses, and why they are technically permissible. Broad AI regulations like those for energy, pharmaceuticals, and other sectors may apply in some form, but ultimately, AI regulation that may work would likely be technical. Physical laws can regulate digital products but would find limitations. There are several AI tools that misuses are possible from that may slip regulations. There are ways that some frontier AI models are self-regulating that may leave little room for adjustments from authorities. Even in the event of presenting models for safety evaluation before public release, frontier AI companies may have done some work, covering what they expect would be sought. How can AI regulation technically fight misuse? This includes the possibility to explore real-time multimodal misuses, across AI models. How is it possible to track and monitor some outputs of all AI tools that are available in a jurisdiction? California Senate has SB 1047 to regulate AI about "Giving companies more flexibility in how they meet their responsible development obligations while still holding them liable if irresponsible behavior leads to catastrophic harm. Allowing a state agency to change the compute threshold for a model to be covered under the bill starting in 2027, rather than keeping it fixed in statute at 10^26 flops. The requirement that the model cost at least $100 million to train cannot be changed. Requiring companies get third-party safety audits by 2028. Strengthen whistleblower protections. Requiring developers of such large “frontier” AI models to take basic precautions, such as pre-deployment safety testing, red-teaming, cybersecurity, safeguards to prevent the misuse of dangerous capabilities, and post-deployment monitoring. " California Senate SB 1047 to regulate AI The European Commission in AI Act enters into force , states that "On 1 August 2024, the European Artificial Intelligence Act (AI Act) enters into force. The Act aims to foster responsible artificial intelligence development and deployment in the EU. Proposed by the Commission in April 2021 and agreed by the European Parliament and the Council in December 2023, the AI Act addresses potential risks to citizens’ health, safety, and fundamental rights. It provides developers and deployers with clear requirements and obligations regarding specific uses of AI while reducing administrative and financial burdens for businesses. The AI Act introduces a uniform framework across all EU countries, based on a forward-looking definition of AI and a risk-based approach:" European Commission in AI Act enters into force The problem with AI regulation is not that laws are not possible around it or that industry standards cannot be set. It is just that the ability to catch misuses on the go is better than regulation at the gate. Simply, it is vital to catch misuses, which in this case goes beyond the aerial view of frontier companies or start-ups but to tools, across usage sources, consistently. This will be like fighting digital piracy at any source it is available within a jurisdiction, not just on major platforms, or pointing to physical laws when digital evasion is knavish. There are research options like penalty-tuning and web scraping of available AI tools . Penalty-tuning could be a way to penalize LLMs, by compute, usage, or language, in a way that the model can know, if it outputs misinformation or deepfakes. penalty-tuning and web scraping of available AI tools Web scraping could be possible for AI tools that are available in search results so that some of the keywords around misuses can be tracked to expect them at their likely destinations, to prevent the harm they might cause. AI regulation is a problem of research first, then of laws. Exploring how to advance, technically, against misuses could guide what directions laws may take. There is a recent preprint on arXiv , Are Emergent Abilities in Large Language Models just In-Context Learning? , where the authors wrote, "We present a novel theory that explains emergent abilities, taking into account their potential confounding factors, and rigorously substantiate this theory through over 1000 experiments. Our findings suggest that purported emergent abilities are not truly emergent, but result from a combination of in-context learning, model memory, and linguistic knowledge. arXiv Are Emergent Abilities in Large Language Models just In-Context Learning? Our work is a foundational step in explaining language model performance, providing a template for their efficient use and clarifying the paradox of their ability to excel in some instances while faltering in others. Thus, we demonstrate that their capabilities should not be overestimated." AI is simply not an existential risk because of prediction-based LLMs. In the human mind , what is called prediction is a type of distribution or relay. LLMs have a substantial part of that, but they do not have what follows after [the incoming half of electrical signals, conceptually]. In the human mind LLMs have a substantial part of that, but they do not have what follows after As AI gets better, with non-concept features and distributions similar to how relays within human memory result in human intelligence , they may pose higher levels of risks. how relays within human memory result in human intelligence

California AI Safety|EU Regulation: LLMs Emergent Abilities and Existential Threat

How to Back Fill Data in An Existing Database

Tether’s success has always attracted people who wanted to prove that the issuer of the world’s leading stablecoin did something wrong. Recent reports indicated that a years-long lawsuit against Tether and the crypto exchange Bitfinex was brought back to life by filing an amended complaint, which accuses them of manipulating crypto prices. In short, the plaintiffs claim that Tether prints USDT without any collateral. Then, the stablecoin is used to buy crypto, including BTC, to artificially boost prices. According to plaintiffs, crypto investors overpaid for crypto because of the pump and, therefore, suffered billions of dollars of damage. The case has been dragging on since 2019, and the recent complaint is aimed to revive it. While the litigation continues, we will look at whether it is theoretically possible to “pump BTC using the uncontrolled emission of USDT.” The current USDT market cap exceeds $114 billion. Reserves, as reported by Tether and audited by BDO Italia, mostly consist of Treasury bills (79.88%), overnight reverse repurchase agreements (12.2%), and money market funds (6.77%). Since 2019, no one has been able to provide evidence supporting claims that Tether was printing USDT out of thin air without respective collateral. Given the attention Tether gets from its competitors who dream about taking its market share, it’s hard to imagine that the company would have been able to get away with such tricks without getting discovered. Taking a look at the chart of USDT market cap in recent years, we can see a fairly strong pullback from $82 billion to $65 billion. This pullback occurred back in 2022 when crypto markets found themselves under pressure amid the Terra (LUNA) disaster and FTX collapse. Such a pullback is hardly surprising as the USDT market cap depends on the demand for stablecoins, which, in turn, is correlated with demand for crypto in general. It should be noted that the correlation between the USDT market cap and demand for “true crypto” is not full. Demand related to purchases of coins and tokens is not the only source of demand for USDT. The stablecoin is actively used for cheap transactions, including international ones. With a daily trading volume that is often close to $70 billion, Tether is the most traded asset in the crypto universe. In addition to transactions, people who reside in countries with weak economies use Tether for savings, which also creates demand for the stablecoin. In order to “pump BTC”, Tether first needs to get away with printing USDT without sufficient collateral (a charge which has not been proven despite multiple attempts) and then use it to secretly buy BTC (obviously, with the help of some shady crypto exchange – another charge which has not been proven). Put simply, it’s just another conspiracy theory. There are three main pillars for USDT demand: crypto demand, transactions demand, and inflation protection demand. If demand grows, Tether is able to purchase collateral and print additional USDT. There’s no need to “pump BTC” with USDT: if demand for Bitcoin grows, demand for USDT will grow, and traders will need USDT to buy BTC on exchanges. Tether’s success has always attracted people who wanted to prove that the issuer of the world’s leading stablecoin did something wrong. Tether’s success has always attracted people who wanted to prove that the issuer of the world’s leading stablecoin did something wrong. Recent reports indicated that a years-long lawsuit against Tether and the crypto exchange Bitfinex was brought back to life by filing an amended complaint, which accuses them of manipulating crypto prices. Recent reports indicated that a years-long lawsuit against Tether and the crypto exchange Bitfinex was brought back to life by filing an amended complaint, which accuses them of manipulating crypto prices. indicated indicated In short, the plaintiffs claim that Tether prints USDT without any collateral. Then, the stablecoin is used to buy crypto, including BTC, to artificially boost prices. According to plaintiffs, crypto investors overpaid for crypto because of the pump and, therefore, suffered billions of dollars of damage. The case has been dragging on since 2019, and the recent complaint is aimed to revive it. While the litigation continues, we will look at whether it is theoretically possible to “pump BTC using the uncontrolled emission of USDT.” The current USDT market cap exceeds $114 billion. Reserves, as reported by Tether and audited by BDO Italia, mostly consist of Treasury bills (79.88%), overnight reverse repurchase agreements (12.2%), and money market funds (6.77%). Since 2019, no one has been able to provide evidence supporting claims that Tether was printing USDT out of thin air without respective collateral. Given the attention Tether gets from its competitors who dream about taking its market share, it’s hard to imagine that the company would have been able to get away with such tricks without getting discovered. Taking a look at the chart of USDT market cap in recent years, we can see a fairly strong pullback from $82 billion to $65 billion. This pullback occurred back in 2022 when crypto markets found themselves under pressure amid the Terra (LUNA) disaster and FTX collapse. Such a pullback is hardly surprising as the USDT market cap depends on the demand for stablecoins, which, in turn, is correlated with demand for crypto in general. It should be noted that the correlation between the USDT market cap and demand for “true crypto” is not full. Demand related to purchases of coins and tokens is not the only source of demand for USDT. The stablecoin is actively used for cheap transactions, including international ones. With a daily trading volume that is often close to $70 billion, Tether is the most traded asset in the crypto universe. In addition to transactions, people who reside in countries with weak economies use Tether for savings, which also creates demand for the stablecoin. In order to “pump BTC”, Tether first needs to get away with printing USDT without sufficient collateral (a charge which has not been proven despite multiple attempts) and then use it to secretly buy BTC (obviously, with the help of some shady crypto exchange – another charge which has not been proven). Put simply, it’s just another conspiracy theory. There are three main pillars for USDT demand: crypto demand, transactions demand, and inflation protection demand. If demand grows, Tether is able to purchase collateral and print additional USDT. There’s no need to “pump BTC” with USDT: if demand for Bitcoin grows, demand for USDT will grow, and traders will need USDT to buy BTC on exchanges.

Can Tether Pump BTC Through Uncontrolled Emission?

Stories, News & Mentions about EBIS

Ebis’s Stories on HackerNoon