With Pillow, digital asset customers are able to invest in one of Pillow's curated, actively managed strategies with a single click! Pillow abstracts away the complexity of Web3 custody management, gas fees, cross-chain bridging, etc to create a simple investing experience for the end-user

Pillow

Anyone who has used galvo-scanners knows that you can't just take and draw a figure without correcting distortion. Distortions are mainly of two types: "barrel" and "pillow". For beginners who are just starting to assemble a device based on a two-axis galvo scanner, I will explain: The galvoscanner consists of two mirrors fixed on the X and Y axes. Due to the fact that the laser beam does not hit the second mirror at an even angle, distortions occur on the surface. The farther the beam is from the center, the more distortion will be. It seems to be a simple geometry and these distortions are a common occurrence, but when you first worked with the galvoscanner, it was not obvious. I thought that the manufacturers of the galvoscanner should have solved this problem. I thought that the manufacturer built a control board inside the galvoscanner that corrects all distortions. I was wrong. And I had to solve this problem. How it all started: It started with the fact that we decided to build an SLS 3D printer. We bought: CO2 laser 50W galvoscanner RC1001 F-theta lens SL1-10.6-210-300Z-LP Differential Line Driver DS26LS31 Teensy 4.1Controller We connected all this and loaded the opengalvo / OPAL firmware https://github.com/opengalvo/OPAL Launched a simple GCODE with a rectangle. But the rectangle didn't work... Then, we made a test GCODE with rectangles of different sizes in order to figure out what is the cause of the distortion.Here is what the galvoscanner drew: After experimenting, I came to the conclusion that there are only two types of distortion. The first type of distortion: “pillow”, is created by a galvoscanner: The second type of distortion, "barrel", is created by the F-theta lens.The distortion is not large, 1mm/200mm, but this is critical for 3D printing. Solve the problem On the topic of distortion, there is a simple and understandable article, the authors of which are Dongyun Wang, Qiwei Yu, Xinpiao Ye https://www.sciencedirect.com/science/article/abs/pii/S0030399213003587 The authors of the article used the following formula to correct the distortion of the image obtained by scanning with a galvoscanner: In our case, we need to do the opposite. The image is applied to the surface using a galvoscanner. We know the coordinates of the image, but we do not know the correct angles of rotation of the mirrors. We need to calculate the angle of rotation of the galvoscan mirrors. More precisely, we need to calculate the angle of rotation of the mirror X.Because on the X axis there is a “pillow” distortion. Our formula will look like this: In C++ the code would look like this: x_angle = atan(x_cord / (sqrt(pow(d,2)+pow(y_cord,2))+e)); Where: - is the angle to rotate the X mirror. x_angle - is the distance from the X-axis of the galvo scanner to the Y-axis e -  is the distance from the Y-axis of the galvo scanner to the surface on which the image is applied d - is the coordinate of the drawn image by X axis (relative to the center) x_cord - is the coordinate of the drawn image by Y axis (relative to the center) y_cord After recalculating the angle of rotation, the “cushion” distortions disappeared: Now we need to remove the barrel distortion on the top and bottom of the rectangle. These distortions are due to the F-theta lens. To correct “barrel” distortion, we can use the same formula:** ** In C++, the code would look like this: y_angle = atan(y_cord / (sqrt(pow(d,2)-pow(x_cord,2)/5))); Where: - is the angle to rotate the X mirror. x_angle - is the coefficient that varies depending on the type of F-theta lens, I picked this coefficient 5 -  is the distance from the Y-axis of the galvo scanner to the surface on which the image is applied d - is the coordinate of the drawn image by X axis (relative to the center) x_cord - is the coordinate of the drawn image by Y axis (relative to the center) y_cord And here is the final result: Finally, a video of the laser work: https://youtu.be/1y2Opalt8_0 This code does not claim high performance, but for our task, it is enough. If you need high speed, then based on this code, you can generate a matrix of coordinate shifts. And when getting coordinates, adjust the values based on the matrix. You can follow the development of the SLS 3D printer project on the fork in GitHub: https://github.com/sinenko/SLS_teensy4_1

Galvo Scanner - Pillow and Barrel Distortions Correction

is that is free and open-source an additional library for the programming language that adds support for , , and in a variety of extension. Pillow Python Imaging Library Python opening manipulating saving Let’s get started The most important class in the is the class, defined in the module with the same name. Python Imaging Library Image We use to image in our as shown below; open ( ) open local directory >>> PIL Image
>>> sample = Image.open( ) from import 'sample.jpg' That’s simple, you have already read an image with , you can now do something with it, You can also check the type of the image we just loaded; pillow image processing >>> type(sample)
< ' . . '> class PIL JpegImagePlugin JpegImageFile You can view your such , , and as shown below image property format size color mode >>> sample.format >>> sample.size
( , )
>>> sample.mode 'JPEG' 250 283 'RGB' Also, you can the you have just read to the screen, using the show method display image >>> sample.show() Output Image Format Conversion Once you are done opening your image with the in one format you can save it again in other formats like from to and much more. pillow library jpg png For instance, let’s try to write a simple Python program to on your project directory which are in format to format. convert all images jpg png os sys PIL Image
jpg_images = [image image os.listdir() image.endswith( )] jpg_image jpg_images: :
        new_name = jpg_image.split( )[ ] + Image.open(jpg_image).save(new_name)
    except IOError error:
        print( .format(jpg_image)) import import from import for in if '.jpg' for in try '.' 0 '.png' as 'Couldn\'t read {} ' Once you run the above code on a project directory consisting of images in jpg format, it gonna open all of them and convert into png just as shown below, you could repeat the same process for other conversions. Image Cropping can also be used to perform whereby you can a of a given image by specifying of the sub rectangle. Pillow image cropping extract sub rectangle coordinates Therefore before you perform , you might wanna know the area of the rectangle to get a of coordinates you would like crop, you can easily observe them by showing to the screen. cropping pair >>> PIL Image
>>> picture = Image.open( )
>>> cord = ( , , , )
>>> new_picture = picture.crop(cord)
>>> new_picture.show() from import 'sample.png' 50 50 300 300 Output As we can see the image has been successful cropped. The coordinates of the surface to be cropped are represented by a . diagonal coordinates Whereby the first two-point is from the upper left diagonal point and next two-point are also the diagonal point from the bottom right. (x, y) (x2, y2) Geometrical Transformation With we can perform some transformation on our image include image size and rotations. pillow geometrical resizing This knowledge plays a great role when for by transforming one image to form other tons of images from different angles. generating data deep learning Image resizing >>> PIL Image
>>> image = Image.open( )
>>> resized_image = image.resize(( , ))
>>> resized_image.save( ) from import 'sample.png' 200 200 'resized.png' When you run the above code you should see a newly resized image on your local directory with a dimension of . 200 x 200 Image Rotation >>> PIL Image
>>> image = Image.open( )
>>> rotated_img = image.rotate( )
>>> rotated_img.save( ) from import 'sample.png' 80 'rotated_img.png' Use to generate of the picture at different angles, doing this way will help us generate as many as data which you can potentially use to train your . rotation feature 360 images same deep learning model Image generator.py PIL Image 
images = [ ] img images: :
        org_img = Image.open(img) angle range( , ):
            image_name = str(angle)+ new_img = org_img.rotate(angle)
            new_img.save(image_name)
    except IOError:
        print( .format(img)) from import 'sample.jpg' for in try for in 1 361 '.jpg' 'Couldn\'t read {}' After running the above script, you should see images of the same original image at different angles as shown below. 360 Image Filtering is a technique for or an image. For example, you can an image to certain features or remove other features. Filtering modifying enhancing filter emphasize is useful for many applications including , , , and . Image filtering smoothing sharpening removing noise edge detection There many filters available on Pillow Library including BLUR, BoxBlur, CONTOUR, FIND_EDGES, Filter, GaussianBlur, Kernel, MaxFilter, MedianFilter, SHARPEN, SMOOTH and etc. Example of Usage. Let’s try finding edges in the below picture using the filter FIND_EDGES Sample image >>> PIL Image
>>> PIL Image, ImageFilter
>>> image = Image.open( )
>>> edges = image.filter(ImageFilter.FIND_EDGES)
>>> edges.show() from import from import 'brain.jpeg' Output Same wise you can experiments with different in Python depending on what you’re trying wanna do. filters Reading Image From open file Also, you can use a pillow to read an Image from a Python file object as shown below; >>> PIL Image
>>> image = Image.open(open( , )) from import 'brain.jpeg' 'rb' Reading Image from URL On this you will have to use Pillow in combination with requests, requests will have to send a get to the server to get the of image and the pillow will read those bytes. requests raw bytes >>> requests
>>> PIL Image
>>> url = >>> raw = requests.get(url, stream=True).raw
>>> Image.open(raw).show() import from import 'https://i1.wp.com/kalebujordan.com/wp-content/uploads/2020/04/code-1084923.png' Output Creating new Images With Pillow you can also create a new blank image, of you might need for a variety of purposes, use to generate an entirely new image. Image.new ( ) Syntax = Image.new(mode, shape, color) new Example of Usage : >>> PIL Image
>>> new_img = Image.new( , ( , ), )
>>> new_img.show() from import 'RGB' 500 500 'blue' Drawing Rectangles on Images Pillow can also be used to draw Rectangle on Images, this commonly applied on object detection whereby you can draw a rectangular box over a detected object. Example of Usage Let’s try drawing a rectangular box inside a blank image >>> PIL Image, ImageDraw
>>> new_img = Image.new( , ( , ), )
>>> pencil = ImageDraw.Draw(new_img)
>>> pencil.rectangle(( , , , ), fill = )
>>> new_img.show() from import 'RGB' 400 400 'black' 200 50 300 300 'green' Output The first two coordinates represent (x,y ) of the left upper diagonal and the next two (x2, y2) represent the coordinate point of the bottom right. Drawing Text on Images We can also use Pillow Library to draw Text on Images PIL Image , ImageDraw, ImageFont new_img = Image.new( , ( , ), ) font = ImageFont.load_default() pencil = ImageDraw.Draw(new_img) pencil.text(( , ), ,  font=font, fill= ) new_img.show() >>> from import >>> 'RGB' 200 200 'black' >>> >>> >>> 100 100 'Hello World' 'blue' >>> Output Well that's all for today, hope you find it useful Now don't be shy, share it with your fellow developer The can be found on kalebujordan.com Original Article In case of any suggestion, comment, or difficulty drop it in community and I will get back to you ASAP.

Intro to Image Processing in Python with Pillow

Building Your AI Radiologist: A Fun Guide to Creating a Pneumonia Detector with VGG16

আপনার এআই রেডিওলজিস্ট তৈরি করা: VGG16 দিয়ে নিউমোনিয়া ডিটেক্টর তৈরির জন্য একটি মজার গাইড

Aufbau Ihres KI-Radiologen: Eine unterhaltsame Anleitung zum Erstellen eines Lungenentzündungsdetektors mit VGG16

Construyendo su radiólogo con IA: una guía divertida para crear un detector de neumonía con VGG16

Construire votre radiologue IA : un guide amusant pour créer un détecteur de pneumonie avec VGG16

अपना AI रेडियोलॉजिस्ट तैयार करें: VGG16 के साथ न्यूमोनिया डिटेक्टर बनाने के लिए एक मजेदार गाइड

AI 放射線科医の育成: VGG16 を使った肺炎検出器の作成ガイド

AI 방사선 전문의 구축: VGG16을 사용하여 폐렴 감지기 생성을 위한 재미있는 가이드

Construindo seu radiologista de IA: um guia divertido para criar um detector de pneumonia com VGG16

Создание вашего искусственного радиолога: забавное руководство по созданию детектора пневмонии с помощью VGG16

Yapay Zeka Radyoloğunuzu Oluşturma: VGG16 ile Pnömoni Dedektörü Oluşturmak İçin Eğlenceli Bir Kılavuz

Xây dựng bác sĩ X quang AI của bạn: Hướng dẫn thú vị để tạo máy phát hiện bệnh viêm phổi với VGG16

打造你的 AI 放射科医生：使用 VGG16 创建肺炎检测器的有趣指南

How Artificial Intelligence Can Make Our Smart Homes, Smarter

কীভাবে কৃত্রিম বুদ্ধিমত্তা আমাদের স্মার্ট হোমগুলিকে আরও স্মার্ট করে তুলতে পারে

Wie künstliche Intelligenz unsere Smart Homes intelligenter machen kann

Cómo la inteligencia artificial puede hacer que nuestras casas inteligentes sean más inteligentes

Comment l’intelligence artificielle peut rendre nos maisons intelligentes plus intelligentes

आर्टिफिशियल इंटेलिजेंस कैसे हमारे स्मार्ट घरों को और भी स्मार्ट बना सकता है

人工知能がスマートホームをさらにスマートにする方法

인공 지능이 스마트 홈을 더욱 스마트하게 만드는 방법

Como a inteligência artificial pode tornar nossas casas inteligentes mais inteligentes

Как искусственный интеллект может сделать наши умные дома умнее

Yapay Zeka Akıllı Evlerimizi Nasıl Daha Akıllı Hale Getirebilir?

Trí tuệ nhân tạo có thể làm cho ngôi nhà thông minh của chúng ta trở nên thông minh hơn như thế nào

人工智能如何让我们的智能家居更加智能

Many organizations that we’ve spoken to are in the exploration phase of using for AI-powered personalization, recommendations, semantic search and anomaly detection. The recent and astronomical improvements in accuracy and accessibility of large language models (LLMs) including BERT and OpenAI have made companies rethink how to build relevant search and analytics experiences. vector search In this blog, we capture engineering stories from 5 early adopters of vector search- Pinterest, Spotify, eBay, Airbnb and Doordash- who have integrated AI into their applications. We hope these stories will be helpful to engineering teams who are thinking through the full lifecycle of vector search all the way from generating embeddings to production deployments. What is vector search? Vector search is a method for efficiently finding and retrieving similar items from a large dataset based on representations of the data in a high-dimensional space. In this context, items can be anything, such as documents, images, or sounds, and are represented as vector embeddings. The similarity between items is computed using distance metrics, such as or , which quantify the closeness of two vector embeddings. cosine similarity Euclidean distance The vector search process usually involves: : Where relevant features are extracted from the raw data to create vector representations using models such as , or Generating embeddings word2vec BERT Universal Sentence Encoder : The vector embeddings are organized into a data structure that enables efficient search using algorithms such as or Indexing FAISS HNSW : Where the most similar items to a given query vector are retrieved based on a chosen distance metric like cosine similarity or Euclidean distance Vector search To better visualize vector search, we can imagine a 3D space where each axis corresponds to a feature. The time and the position of a point in the space is determined by the values of these features. In this space, similar items are located closer together and dissimilar items are farther apart. GITHUB JULIE-MILLS ^
  |          x Item 1
  |         /
  |        /
  |       /x Item 2
  |      /
  |     /
  |    /x Item 3
  |   /
  |  /
  | /
  +-------------------> Given a query, we can then find the most similar items in the dataset. The query is represented as a vector embedding in the same space as the item embeddings, and the distance between the query embedding and each item embedding is computed. The item embeddings with the shortest distance to the query embedding are considered the most similar. Query item:         x
                    |
  ^                 |
  |          x Item 1|
  |         /       |
  |        /        |
  |       /x Item 2 |
  |      /          |
  |     /           |
  |    /x Item 3    |
  |   /             |
  |  /              |
  | /               |
  +-------------------> This is obviously a simplified visualization as vector search operates in high-dimensional spaces. In the next sections, we’ll summarize 5 engineering blogs on vector search and highlight key implementation considerations. The full engineering blogs can be found below: by Jinfeng Zhuang at Pinterest PinText: A Multitask Text Embedding System in Pinterest by Alexandre Tamborrino at Spotify Introducing Natural Language Search for Podcast Episodes by Senthilkumar Gopal, Shubhangi Tandon, Christopher Miller, Deepika Srinivasan, Rui Kong, Selcuk Kopru and Srinivas Bhagavathula at eBay How eBay’s New Search Feature Was Inspired By Window Shopping by Mihajlo Grbovic at AirBnb Listing Embeddings in Search Ranking by Mitchell Koch, Aamir Manasawala, Raghav Ramesh at Doordash Personalized Store Feed with Vector Embeddings Pinterest: Interest search and discovery Pinterest uses for image search and discovery across multiple areas of its platform, including recommended content on the home feed, related pins and search using a multitask learning model. vector search A multi-task model is trained to perform multiple tasks simultaneously, often sharing underlying representations or features, which can improve generalization and efficiency across related tasks. In the case of Pinterest, the team trained and used the same model to drive recommended content on the homefeed, related pins and search. Pinterest trains the model by pairing a users search query (q) with the content they clicked on or pins they saved (p). Here is how Pinterest created the (q,p) pairs for each task: : Word embeddings are derived from the selected subject (q) and the pin clicked on or saved by the user (p). Related Pins : Word embeddings are created from the search query text (q) and the pin clicked on or saved by the user (p). Search : Word embeddings are generated based on the interest of the user (q) and the pin clicked on or saved by the user (p). Homefeed To obtain an overall entity embedding, Pinterest averages the associated word embeddings for related pins, search and the homefeed. Pinterest created and evaluated its own supervised Pintext-MTL (multi-task learning) against unsupervised learning models including GloVe, word2vec as well as a single-task learning model, PinText-SR on precision. PinText-MTL had higher precision than the other embedding models, meaning that it had a higher proportion of true positive predictions among all positive predictions. Pinterest also found that multi-task learning models had a higher recall, or a higher proportion of relevant instances correctly identified by the model, making them a better fit for search and discovery. To put this all together in production, Pinterest has a multitask model trained on streaming data from the homefeed, search and related pins. Once that model is trained, vector embeddings are created in a large batch job using either Kubernetes+Docker or a map-reduce system. The platform builds a search index of vector embeddings and runs a K-nearest neighbors (KNN) search to find the most relevant content for users. Results are cached to meet the performance requirements of the Pinterest platform. Spotify: Podcast search Spotify combines keyword and to retrieve relevant podcast episode results for users. As an example, the team highlighted the limitations of keyword search for the query “electric cars climate impact”, a query which yielded 0 results even though relevant podcast episodes exist in the Spotify library. To improve recall, the Spotify team used Approximate Nearest Neighbor (ANN) for fast, relevant podcast search. semantic search The team generates vector embeddings using the as it is multilingual, supporting a global library of podcasts, and produces high-quality vector embeddings. Other models were also evaluated including , a model trained on a big corpus of text data, but found that BERT was better suited for word embeddings than sentence embeddings and was pre-trained only in English. Universal Sentence Encoder CMLM model BERT Spotify builds the vector embeddings with the query text being the input embedding and a concatenation of textual metadata fields including title and description for the podcast episode embeddings. To determine the similarity, Spotify measured the cosine distance between the query and episode embeddings. To train the base Universal Sentence Encoder CMLM model, Spotify used positive pairs of successful podcast searches and episodes. They incorporated in-batch negatives, a technique highlighted in papers including and , to generate random negative pairings. Testing was also conducted using synthetic queries and manually written queries. Dense Passage Retrieval for Open-Domain Question Answering (DPR) Que2Search: Fast and Accurate Query and Document Understanding for Search at Facebook To incorporate vector search into serving podcast recommendations in production, Spotify used the following steps and technologies: : Spotify indexes the episode vectors offline in batch using , a search engine with native support for ANN. One of the reasons that Vespa was chosen is that it can also incorporate metadata filtering post-search on features like episode popularity. Index episode vectors Vespa : Spotify uses to generate a query vector. Vertex AI was chosen for its support for GPU inference, which is more cost effective when using large transformer models to generate embeddings, and for its query cache. After the query vector embedding is generated, it is used to retrieve the top 30 podcast episodes from Vespa. Online inference Google Cloud Vertex AI Semantic search contributes to the identification of pertinent podcast episodes, yet it is unable to fully supplant keyword search. This is due to the fact that semantic search falls short of exact term matching when users search an exact episode or podcast name. Spotify employs a hybrid search approach, merging semantic search in Vespa with keyword search in , followed by a conclusive re-ranking stage to establish the episodes displayed to users. Elasticsearch eBay: Image search Traditionally, search engines have displayed results by aligning the search query text with textual descriptions of items or documents. This method relies extensively on language to infer preferences and is not as effective in capturing elements of style or aesthetics. eBay introduces to help users find relevant, similar items that meet the style they’re looking for. image search eBay uses a multi-modal model which is designed to process and integrate data from multiple modalities or input types, such as text, images, audio, or video, to make predictions or perform tasks. eBay incorporates both text and images into its model, producing image embeddings utilizing a Convolutional Neural Network (CNN) model, specifically , and title embeddings using a text-based model such as . Every listing is represented by a vector embedding that combines both the image and title embeddings. Resnet-50 BERT Once the multi-modal model is trained using a large dataset of image-title listing pairs and recently sold listings, it is time to put it into production in the site search experience. Due to the large number of listings at eBay, the data is loaded in batches to HDFS, eBay’s data warehouse. eBay uses Apache Spark to retrieve and store the image and relevant fields required for further processing of listings, including generating listing embeddings. The listing embeddings are published to a columnar store such as HBase which is good at aggregating large-scale data. From HBase, the listing embedding is indexed and served in Cassini, a search engine created at eBay. The pipeline is managed using Apache Airflow, which is capable of scaling even when there is a high quantity and complexity of tasks. It also provides support for Spark, Hadoop, and Python, making it convenient for the machine learning team to adopt and utilize. Visual search allows users to find similar styles and preferences in the categories of furniture and home decor, where style and aesthetics are key to purchase decisions. In the future, eBay plans to expand visual search across all categories and also help users discover related items so they can establish the same look and feel across their home. AirBnb: Real-time personalized listings Search and similar listings features drive 99% of bookings on the AirBnb site. AirBnb built a to improve similar listing recommendations and provide real-time personalization in search rankings. listing embedding technique AirBnb realized early on that they could expand the application of embeddings beyond just word representations, encompassing user behaviors including clicks and bookings as well. To train the embedding models, AirBnb incorporated over 4.5M active listings and 800 million search sessions to determine the similarity based on what listings a user clicks and skips in a session. Listings that were clicked by the same user in a session are pushed closer together; listings that were skipped by the user are pushed further away. The team settled on the dimensionality of a listing embedding of d=32 given the tradeoff between offline performance and memory needed for online serving. https://youtu.be/aWjsUEX7B1I?si=GREROVrWxWqtqL-m&embedable=true AirBnb found that certain listings characteristics do not require learning, as they can be directly obtained from metadata, such as price. However, attributes like architecture, style, and ambiance are considerably more challenging to derive from metadata. Before moving to production, AirBnb validated their model by testing how well the model recommended listings that a user actually booked. The team also ran an A/B test comparing the existing listings algorithm against the vector embedding-based algorithm. They found that the algorithm with vector embeddings resulted in a 21% uptick in CTR and 4.9% increase in users discovering a listing that they booked. The team also realized that vector embeddings could be used as part of the model for real-time personalization in search. For each user, they collected and maintained in real time, using Kafka, a short-term history of user clicks and skips in the last two weeks. For every search conducted by the user, they ran two similarity searches: based on the geographic markets that were recently searched and then the similarity between the candidate listings and the ones the user has clicked/skipped Embeddings were evaluated in offline and online experiments and became part of the real-time personalization features. Doordash: Personalized store feeds Doordash has a wide variety of stores that users can choose to order from and being able to surface the most relevant stores using personalized preferences improves search and discovery. Doordash wanted to apply to its store feed algorithms using vector embeddings. This would enable Doordash to uncover similarities between stores that were not well-documented including if a store has sweet items, is considered trendy or features vegetarian options. latent information Doordash used a derivative of word2vec, an embedding model used in natural language processing, called store2vec that it adapted based on existing data. The team treated each store as a word and formed sentences using the list of stores viewed during a single user session, with a maximum limit of 5 stores per sentence. To create user vector embeddings, Doordash summed the vectors of the stores from which users placed orders in the past 6 months or up to 100 orders. As an example, Doordash used vector search to find similar restaurants for a user based on their recent purchases at popular, trendy joints 4505 Burgers and New Nagano Sushi in San Francisco. Doordash generated a list of similar restaurants measuring the cosine distance from the user embedding to store embeddings in the area. You can see that the stores that were closest in cosine distance include Kezar Pub and Wooden Charcoal Korean Village BBQ. Doordash incorporated store2vec distance feature as one of the features in its larger recommendation and personalization model. With vector search, Doordash was able to see a 5% increase in click-through-rate. The team is also experimenting with new models like , model optimizations and incorporating real-time onsite activity data from users. seq2seq Key considerations for vector search Pinterest, Spotify, eBay, Airbnb and Doordash create better search and discovery experiences with vector search. Many of these teams started out using text search and found limitations with fuzzy search or searches of specific styles or aesthetics. In these scenarios, adding vector search to the experience made it easier to find relevant, and often personalized, podcasts, pillows, rentals, pins and eateries. There are a few decisions that these companies made that are worth calling out when implementing vector search: : Many started out using an off-the-shelf model and then trained it on their own data. They also recognized that language models like word2vec could be used by swapping words and their descriptions with items and similar items that were recently clicked. Teams like AirBnb found that using derivatives of language models, rather than image models, could still work well for capturing visual similarities and differences. Embedding models : Many of these companies opted to train their models on past purchase and click through data, making use of existing large-scale datasets. Training : While many companies adopted ANN search, we saw that Pinterest was able to combine metadata filtering with KNN search for efficiency at scale. Indexing : Vector search rarely replaces text search. Many times, like in Spotify’s example, a final ranking algorithm is used to determine whether vector search or text search generated the most relevant result. Hybrid search : We’re seeing many teams use batch-based systems to create the vector embeddings, given that these embeddings are rarely updated. They employ a different system, frequently Elasticsearch, to compute the query vector embedding live and incorporate real-time metadata in their search. Productionizing Rockset, a real-time search and analytics database, recently added support for . Give vector search on Rockset a try for real-time personalization, recommendations, anomaly detection and more by starting a with $300 in credits today. vector search free trial

A Look Into 5 Use Cases for Vector Search from Major Tech Companies

প্রধান প্রযুক্তি কোম্পানি থেকে ভেক্টর অনুসন্ধানের জন্য 5টি ব্যবহারের ক্ষেত্রে একটি নজর৷

Ein Blick auf 5 Anwendungsfälle für die Vektorsuche bei großen Technologieunternehmen

Una mirada a cinco casos de uso de búsqueda de vectores de las principales empresas tecnológicas

Un aperçu de 5 cas d'utilisation de la recherche de vecteurs par de grandes entreprises technologiques

प्रमुख टेक कंपनियों द्वारा वेक्टर सर्च के लिए 5 उपयोग मामलों पर एक नज़र

大手テクノロジー企業によるベクトル検索の 5 つのユースケースを検証

주요 기술 회사의 벡터 검색에 대한 5가지 사용 사례 살펴보기

Uma análise de 5 casos de uso para pesquisa vetorial de grandes empresas de tecnologia

Взгляд на 5 вариантов использования векторного поиска от крупных технологических компаний

Büyük Teknoloji Şirketlerinden Vektör Aramaya Yönelik 5 Kullanım Durumuna Bir Bakış

Xem xét 5 trường hợp sử dụng tìm kiếm vectơ từ các công ty công nghệ lớn

了解主要科技公司矢量搜索的 5 个用例

This is paper is available on arxiv under CC 4.0 DEED license.
Authors:
(1) Dhruv Shah, UC Berkeley and he contributed equally;
(2)  Michael Equi, UC Berkeley and he contributed equally;
(3) Blazej Osinski, University of Warsaw;
(4) Fei Xia, Google DeepMind;
(5) Brian Ichter, Google DeepMind;
(6) Sergey Levine, UC Berkeley and Google DeepMind. Table of Links Abstract & Introduction
Related Work
Problem Formulation and Overview
LFG: Scoring Subgoals by Polling LLMs
LLM Heuristics for Goal-Directed Exploration
System Evaluation
Discussion and References
A. Implementation Details
B. Prompts A Implementation Details A.1 Hyperparameters A.2 Computational Resources A.3 Real World Results Generating Prompts: For both topological and geometric maps we use hand engineered methods for clustering objects in ways that the LLM can efficiently reason over. For geometric maps we implement two functions: parseObjects and clusterObjects. In our implementation, parseObejcts filters the geometric map and identifies the cluster centers among each class. clusterObjects takes the cluster centers and performs agglomerative clustering with a threshold of 6 meters, which is roughly the size of one section of a standard house. For topological maps we rely on the configuration of the four cameras to automatically perform parsing and clustering. In our implementation all the objects detected in each frame from either the front, left, right, or rear facing cameras is considered a single cluster. Perception: For the hardware, we use a locobot base with a four HD logitech web cameras that are positioned at 90 degrees relative to each other. At each step of LFG each of four cameras is recorded and frames are semantically annotated. LFG directly uses these frames to determine if the robot should continue to move forward, turn left, turn right, or turn around a full 180 degrees. To improve the performance of our system we choose to whitelist a subset of the 20,000 classes. This reduces the size of the API calls to the language models and helps steer the LLM to focus on more useful information. Following is the complete whitelist used in our experiments: • toaster • projector • chair • kitchen table • sink • kitchen sink • water faucet • faucet • microwave oven • toaster oven • oven • coffee table • coffee maker • coffeepot • dining table • table • bathtub • bath towel • urinal • toilet • toilet tissue • refrigerator • automatic washer • washbasin • dishwasher • television set • sofa • sofa bed • bed • chandelier • ottoman • dresser • curtain • shower curtain • trash can • garbage • cabinet • file cabinet • monitor (computer equipment) • computer monitor • computer keyboard • laptop computer • desk • stool • hand towel • shampoo • soap • drawer • pillow Low-level Policy: The low-level policy running on the robot is the NoMaD goal-conditioned diffusion policy trained to avoid obstacles during exploration and determine which frontiers can be explored further [41]. High-level Planning: For real-world experiments, we follow the setup of ViKiNG [34], where the agent runs a simple frontier-based exploration algorithm and incorporates the LLM scores as goaldirected heuristics to pick the best subgoal frontier. For simulation experiments, we use a geometric map coupled with frontier-based exploration, following the setup of Chaplot et al. [38]. Algorithms 3 and 4 summarize the high-level planning module in both cases. A.4 More Experiment Rollouts Figure 6 shows an example where the negative scoring is essential to LFG’s success. Figures 7 and 8 show examples of LFG deployed in a previously unseen apartment and an office building, successfully exploring the environments to find an oven and a kitchen sink. This is paper is available on arxiv under CC 4.0 DEED license. Authors: (1) Dhruv Shah, UC Berkeley and he contributed equally; (2)  Michael Equi, UC Berkeley and he contributed equally; (3) Blazej Osinski, University of Warsaw; (4) Fei Xia, Google DeepMind; (5) Brian Ichter, Google DeepMind; (6) Sergey Levine, UC Berkeley and Google DeepMind. This is paper is available on arxiv under CC 4.0 DEED license. Authors: Authors: (1) Dhruv Shah, UC Berkeley and he contributed equally; (2)  Michael Equi, UC Berkeley and he contributed equally; (3) Blazej Osinski, University of Warsaw; (4) Fei Xia, Google DeepMind; (5) Brian Ichter, Google DeepMind; (6) Sergey Levine, UC Berkeley and Google DeepMind. Table of Links Abstract & Introduction Related Work Problem Formulation and Overview LFG: Scoring Subgoals by Polling LLMs LLM Heuristics for Goal-Directed Exploration System Evaluation Discussion and References A. Implementation Details B. Prompts Abstract & Introduction Abstract & Introduction Related Work Related Work Problem Formulation and Overview Problem Formulation and Overview LFG: Scoring Subgoals by Polling LLMs LFG: Scoring Subgoals by Polling LLMs LLM Heuristics for Goal-Directed Exploration LLM Heuristics for Goal-Directed Exploration System Evaluation System Evaluation Discussion and References Discussion and References A. Implementation Details A. Implementation Details B. Prompts B. Prompts A Implementation Details A.1 Hyperparameters A.2 Computational Resources A.3 Real World Results Generating Prompts: For both topological and geometric maps we use hand engineered methods for clustering objects in ways that the LLM can efficiently reason over. For geometric maps we implement two functions: parseObjects and clusterObjects. In our implementation, parseObejcts filters the geometric map and identifies the cluster centers among each class. clusterObjects takes the cluster centers and performs agglomerative clustering with a threshold of 6 meters, which is roughly the size of one section of a standard house. For topological maps we rely on the configuration of the four cameras to automatically perform parsing and clustering. In our implementation all the objects detected in each frame from either the front, left, right, or rear facing cameras is considered a single cluster. Generating Prompts: Perception: For the hardware, we use a locobot base with a four HD logitech web cameras that are positioned at 90 degrees relative to each other. At each step of LFG each of four cameras is recorded and frames are semantically annotated. LFG directly uses these frames to determine if the robot should continue to move forward, turn left, turn right, or turn around a full 180 degrees. To improve the performance of our system we choose to whitelist a subset of the 20,000 classes. This reduces the size of the API calls to the language models and helps steer the LLM to focus on more useful information. Following is the complete whitelist used in our experiments: Perception: • toaster • projector • chair • kitchen table • sink • kitchen sink • water faucet • faucet • microwave oven • toaster oven • oven • coffee table • coffee maker • coffeepot • dining table • table • bathtub • bath towel • urinal • toilet • toilet tissue • refrigerator • automatic washer • washbasin • dishwasher • television set • sofa • sofa bed • bed • chandelier • ottoman • dresser • curtain • shower curtain • trash can • garbage • cabinet • file cabinet • monitor (computer equipment) • computer monitor • computer keyboard • laptop computer • desk • stool • hand towel • shampoo • soap • drawer • pillow Low-level Policy: The low-level policy running on the robot is the NoMaD goal-conditioned diffusion policy trained to avoid obstacles during exploration and determine which frontiers can be explored further [41]. Low-level Policy: High-level Planning: For real-world experiments, we follow the setup of ViKiNG [34], where the agent runs a simple frontier-based exploration algorithm and incorporates the LLM scores as goaldirected heuristics to pick the best subgoal frontier. For simulation experiments, we use a geometric map coupled with frontier-based exploration, following the setup of Chaplot et al. [38]. Algorithms 3 and 4 summarize the high-level planning module in both cases. High-level Planning: A.4 More Experiment Rollouts Figure 6 shows an example where the negative scoring is essential to LFG’s success. Figures 7 and 8 show examples of LFG deployed in a previously unseen apartment and an office building, successfully exploring the environments to find an oven and a kitchen sink.

Navigation with Large Language Models: Implementation Details

Blockchain has been a mixed bag from the start. A technology that gave birth to revolutionary ideas was quickly exposed to the harsh reality. People were not welcoming, and many could not understand how and why we needed it. But now and then, developers come up with efficient ways to implement the technology and shake up conventional approaches. This is what happened to supply chain management when blockchain solutions were introduced. The Walmart Way Walmart is a household name. A huge supermarket chain dealing with everything from pillows to oranges. It’s not entirely surprising that the company is always looking to polish and modernize its practices. In 2016, Walmart began the development of a solution for tracing grocery products using blockchain. Along with IBM, the company launched the Walmart Food Traceability Initiative. For that to function, suppliers had to include information about each purchase from farmers or manufacturers in the blockchain. But thanks to technology, what used to take hours or even days is now possible in seconds. Retailers could check the origins of products quickly. introduced blockchain when the time came to solve long-standing issues with invoices and payments for freight carriers. The company oversees supply for Walmart stores all over Canada. Each carrier has plenty of information that needs to be reflected in an invoice, including fuel, stops, temperature updates for perishable goods, etc., over 200 data points in general. In such cases, a small discrepancy can cause premature expiration and subsequent financial loss. And to anyone familiar with blockchain it truly seems like a perfect solution. Walmart Canada Walmart Canada partnered with DLT Labs to develop a solution based on distributed ledger technology. The company has deemed it a success, stating that about 1% of invoices have to be disputed (compared to 70% before the implementation). Plenty Of Oil And Fish Walmart might be the biggest adopter of blockchain, but it’s not the only one. Various companies, from diamond sellers to fish manufacturers, distribute blockchain-enabled goods. And, most of them follow the largest retailer’s experience. Abu Dhabi National Oil Company has even a similar IBM-designed tracing system to monitor and manage oil and gas supplies. Once again, transparency is listed as an advantage as the information is available to “everyone who needs to see it.” implemented Needless to say, blockchain could be a fan favorite for businesses big and small. A bunch of buzzwords that increase trust can make for good publicity, and innovation is always a must. The Good And The Bad Not only can blockchain allow for lightning-fast access to any information about products, but it also inspires developers and retailers alike to develop new solutions. Blockchain can remove the need for third parties such as auditors, which lowers the cost of products. A of Walmart’s blockchain experience lists several additional advantages of using distributed ledger technology. case study • A single block can contain information about manufacturers, including labor practices. • Blockchain allows traceability. • Blockchain helps eliminate unnecessary steps in supply chains. • An automated supply chain helps lower the cost of products. But like every other idea that serves humanity, this sounds too good to be true. If anything, we know very well how little the digital world truly correlates with the physical one. First of all, the question of public vs. private blockchain is raised. Most of the aforementioned solutions work on a private ledger (i.e., it’s accessible and changeable by its users only). So the retailer can check the origins of products and make sure everything is in order. But a customer can’t necessarily do the same. Sometimes due to lack of knowledge, others—due to lack of access. If an instance occurs in which the retailer and the supplier find it necessary to withhold information from a customer, the latter is left completely unprotected. Aside from access, we are facing another issue: A public blockchain has thousands of validators, while a private chain has only a few. Transparency for everyone is one of the fundamental principles of blockchain, and the authenticity of any information on a private chain remains questionable. At the end of the day, what tells a customer that the product they are buying is authentic? The level of trust for multimillion conglomerates has probably . And it’s not like every diamond or fish has a special tracking number. None of that protects customers from purchasing an illegally obtained product or forces sellers to avoid unethical manufacturers. On the contrary, the veil of transparency that isn’t quite clear to everyone could be covering more than it shows. never been lower Even if some issues like blockchain-powered tracing of products in real-time are not quite achievable just yet, we should be looking for new possibilities. Speed and accuracy within the process may be a priority for a supplier, but the ultimate goal should be to enable access to every participant, from a farmer in Iowa to a customer in D.C. It’s A Virtual World After All At this time, we likely could benefit way more from involving this technology in the management of digital or virtual goods: software, digital items and even working hours. The latter opens up curious possibilities in the current environment. For instance, a company that is under sanctions closes its HQ in one country and opens a proxy HQ in another. While management may have relocated, casual workers remain in their original location, working under a VPN. If we employ a blockchain solution and distribute working hours through a public blockchain, circumvention of sanctions would be harder, if not impossible. Blockchain delivers the most necessary element for any supply chain—unalterable data storage. Still, we could benefit way more from involving this technology in the management of digital or virtual goods. And that’s something to work on. Follow me on . Check out my . LinkedIn website

Why Blockchain Falls Short In Supply Chain Management

Do you ever find yourself whispering these excuses to yourself? "When I finally get this promotion, THEN I'll have time to start my side hustle." "When the kids are a little older, THEN I'll finally get back to painting." "When I have a bigger house, THEN I can start hosting those dinner parties I've dreamed of." "When I have more money saved, THEN I'll feel secure enough to travel." If you do, you're not alone. I've been there too.  We all get stuck in the trap of thinking the perfect conditions will appear like magic down the road. Only then, the stars aligning, will we somehow be ready to chase our dreams/do the thing. But here's the spoiler no one likes to admit: That perfect "X" moment will NEVER arrive. There will ALWAYS be something—a new project at work, a sick kid, an unexpected car repair—demanding your attention and energy. "Someday" is the biggest lie we tell ourselves. It's a fluffy pillow to soften the blow of our inaction.  We wrap ourselves in its comforting excuse, but deep down, we know that if we don't make a move , "someday" will fade into "never."  Worse, it'll fade into a nagging regret you carry with you for years to come. now The Truth About Readiness (And Why it's Not What You Think) Here comes the hard-to-swallow truth: Readiness, in the way we typically picture it, is a complete myth. You don't magically become infused with expertise, bulletproof confidence, and infinite resources after some external factor changes. Think about it this way: Did Beyoncé wake up one day feeling flawlessly ready to conquer the world, or did she build her skills, stage presence, and business savvy through countless hours of sweat and imperfect performances? Remember that kid in your high school class who seemed effortlessly good at everything? The one who aced tests without studying and charmed even the grouchiest teachers?  I bet if you dug deeper, you'd learn they were putting in work behind the scenes when no one was looking. Readiness isn't bestowed;  it's BUILT. Each imperfect step you take forges that readiness in the fire of action. The "Start Now" Mentality: A Personal Story Years ago, I desperately wanted to leave my corporate job and become a writer (start a newsletter… blog… anything).  But "someday" was my best friend. Someday, I'd have the perfect freelance client roster. Someday, I'd finally figure out this whole online marketing thing. Someday... Paralysis set in. All I did was watch videos, buy fancy courses, and dream.  One day, in a moment of frustration with myself, I decided to try something new. I forced myself to write a single blog post and hit publish...even though I thought it was terrible. YouTube The next day, I did it again. And again. Some were cringe-worthy, and some gained a bit of traction. But it didn't matter.  The act of doing, however messy, was chipping away at my fear and building my skills.  Within a year, I had enough momentum and income to make the leap into earning some income, selling ads on a newsletter (fast forward to today, the newsletter has 320,000 subscribers). That never happens if I keep waiting for the magic moment of readiness. Messy Action > Perfect Inaction Let's make this real. Let's say your dream is to get in shape finally. Here's the old mindset vs. the action-first mindset: Old "Someday" Mindset: I need to join the fancy gym. I need the perfect workout clothes. I need to find a trainer and get a program designed... Result:  You end up spending weeks comparing gym memberships and end up doing  nothing. Action-First Mindset: It's rainy out, and I'm tired, but my old sneakers are fine. I'll do one of those YouTube workout videos in my living room. It might be awkward, but I'm gonna get moving. Result: It's not glamorous, but it's a start! You're BUILDING the habit of moving your body regularly. Each day it gets a tiny bit easier. "But What About..." (Common Barriers Busted) I can hear the objections whispering in your head. So let's tackle some of the most common ones: Truth bomb: We ALL have the same 24 hours. Successful people aren't gifted extra time; they get ruthless about priorities.  Audit your day. Where does your time vanish on social media, mindless TV, or tasks that could be outsourced? Can you carve out even 15 minutes a day for your goal? You'll be shocked at what you can do with focused intention. "I don't have TIME!" Aren't you more afraid of never trying and living with regret? Failure is inevitable on the path to anything worthwhile. Embrace it as a teacher. Each stumble informs what to do differently –  that's growth happening. "I'm afraid of FAILING!" This is where creativity is your superpower. Can't afford a class? There's an insane amount of free knowledge online. Need equipment?  Look for secondhand options or see if a friend will loan it.  Remember, it's about your resourcefulness, not the size of your wallet. "I don't have the RESOURCES!" The Bottom Line: "Someday" Will Betray You Every day spent waiting for perfect conditions is a day you're not actively building the life you truly want. Imagine yourself five years from now. Do you want to look back and see a trail of "somedays" fading into the distance? Or do you want to see how far you've come because of those imperfect first steps you finally took? The "Start Now" Challenge (Take it Seriously!) I want to challenge you, and I mean challenge you, to take one small, uncomfortable step toward your goal today.  Not tomorrow, not after you research/plan/organize more – TODAY. Don't let yourself off the hook! really Here's how: What's the absolute smallest action that you've been avoiding? (Ex: write 50 words, send a sales email, register for that one class, do 10 pushups) The Tiny Task: Don't overcomplicate it. Clear your space, set a timer for 20 minutes, do whatever gets that task in front of you. Remove the Barriers: Don't judge the quality, just focus on getting it DONE. Prove to yourself you can follow through. DO IT (no matter how imperfectly): You Are Readier Than You Think You wouldn't be reading this if there wasn't a fire inside of you, a voice whispering that you're meant for more.  Don't betray that part of yourself. Take the leap today, even just a tiny one. One step leads to another and then another, and suddenly that "someday" dream is within your grasp. The choice is yours. Are you ready to ditch the excuses and finally start building a life filled with purpose and passion? Then it's time to stop waiting on "X" and create your readiness. Your future self will thank you. Have a great week! Scott Also published . here

Choosing Messy Action Over Perfect Inaction

PILLOW

pillow.fund

-- employees

Since n.d.

COMPANY RANKING

PILLOW

EVERGREEN INDEX #8676

Representing this company? Click here to claim and customize this page!

Pillow WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!

PILLOW

pillow.fund

-- employees

Since n.d.

COMPANY RANKING

PILLOW

EVERGREEN INDEX #8676

RELATED COMPANIESRANK

STORIES ABOUT PILLOW

Representing this company? Click here to claim and customize this page!

Pillow WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!