#pgvector

[govindhtech]

NeuroPace Enhances iEEG Data Analysis with AlloyDB Omni

What is iEEG

Mountain View-based NeuroPace, Inc. makes the FDA-approved RNS System responsive neurostimulator device for people with refractory focal onset epilepsy. A lead with four electrode connections can be connected to a neurostimulator in up to two ways. Both the delivery of electrical pulses and the detection of aberrant patterns unique to each patient are controlled by physicians. Intracranial electroencephalograms, or iEEGs, are recorded by the device and span four channels with a sample rate of 250 Hz each. Typically, the recordings contain roughly 90 seconds of data. Almost 5,000 patients have provided over 16 million iEEG files to date.Image Credit to Google Cloud

Determining efficient stimulation patterns for the reduction of seizures is a primary research objective at NeuroPace. A theory posits that treatment settings that have proven successful for patients with comparable iEEG activity may also work for newly diagnosed patients or patients requiring modifications to their current stimulation regimen, thus potentially enhancing the traditional trial-and-error method of determining stimulation programming settings.

Physicians can swiftly identify similar patient profiles based on chosen iEEG files if large-scale iEEG data is searched for similar brain activity patterns. This is necessary for such data-driven procedures to be feasible. Previously, locating similar iEEG files among patients necessitated a convoluted processing pipeline that included grouping iEEG data within patients, locating cluster centres, and locating approximate nearest neighbours (ANN) from other patients using dimensionality reduction techniques like PCA and t-SNE. The flexibility and practical utility of the programme were limited because the approximate nearest neighbours were only computed once every several months using a small sample of new patient iEEG recordings.

Positively, vector databases may now be queried directly for comparable vector embeddings thanks to recent developments. Without having to complete clustering procedures beforehand, this invention may allow doctors to choose any iEEG file from a patient and locate comparable cross-patient iEEG data. As fresh iEEG files become available, all that is needed is to keep the vector database updated. Improved scalability could result from this reduction, which would make it much easier to query comparable iEEG files over millions of entries.

Using embedding models to turn iEEG data into vectors, the NeuroPace  AI team and Google Cloud developers carried out a proof-of-concept study. The vector data was then stored in a vector database called AlloyDB for PostgresQL. AlloyDB is a fully managed database that performs vector similarity searches based on the pgvector extension and is compatible with PostgreSQL, making it ideal for heavy transaction workloads.

AlloyDB Omni

AlloyDB Omni, a version of the database that can be downloaded and used anywhere, further permits on-premises hosting of the database, preserving the data inside the confines of an on-prem HIPAA-compliant environment. Reducing reliance on external network connectivity by having the database on-premise also lessens the chance of application outages that might arise from hosting the database externally while the remainder of the application is hosted on-premise.

Google Cloud handled almost 1.2 million iEEG files from 414 clinical trial participants in this proof-of-concept project. 20 patients’ worth of data were used for testing, and 394 patients’ data were added to the AlloyDB cloud service. A unique embedding model created by the NeuroPace  AI team converted each iEEG file into a spectrogram image and then into vectors. After that, 50 randomly chosen iEEG files from the test cohort were utilised to query the AlloyDB vector database, which now contained these vectors (Figure 2).Image credit to Google Cloud

AlloyDB with PGvector offers three distinct index types (Hierarchical Navigable Small World (HNSW), IVFFLAT, and IVF) that help reduce latency while conducting similarity searches in comparison to a brute-force search:

Using a graph-based technique, HNSW creates several layers of interconnected nodes to create more effective search pathways even for big datasets.

The {IVFFLAT} index balances speed and accuracy by first grouping vectors into coarse groups using a tree-based clustering technique, and then conducting a more thorough search inside the most comparable clusters.

Google AlloyDB Omni

A recent addition to AlloyDB  AI enhancements is the new “IVF” index, which increases the total number of dimensions supported per vector and dramatically reduces query time by utilising deeper integrations with AlloyDB query processing in addition to Google quantization techniques.

In actuality, distinct indices (as well as the corresponding algorithms) can perform very differently under very diverse use situations. Google  Cloud conducted a detailed benchmarking across IVF and HNSW indexes for the NeuroPace use case of locating similar cross-patient iEEGs. Recall, or the percentage of results in brute-force queries, and latency, or how quickly the query could finish, were both measured.

An analysis of the performance of these two approximate nearest neighbour (ANN) algorithms shows that IVF has high recall rates (~0.9) and a median query latency of roughly 60 ms, while HNSW indexing performed slightly worse (0.8) and was slower (median query latency of 160 ms) than IVF. To balance speed and latency, both indexes provide a variety of characteristics.

While brute force took about 14.7 seconds to find comparable iEEG data, both approaches performed noticeably better in terms of query time. The histogram of the recall and query latency for the two distinct indexing strategies in comparison to brute force. The output of a single sample iEEG query file from a test patient is displayed .

These results excite NeuroPace, since they could further the research towards effectively navigating large amounts of iEEG data. The development of algorithms that help determine the best programming settings for the RNS System may be made possible by this breakthrough. The new ScaNN index from AlloyDB may potentially help to further enhance usability and performance, and Google Cloud is eager to test it out.

Read more on govindhtech.com

[quotejungle]

AI アプリケーションの��築に大金を費やす必要はありません。最高の AI 開発ツールはオープンソースであり、AI を誰もが利用できる優れたエコシステムが進化しています。 このオープンソース AI スタックの主要コンポーネントは次のとおりです。 1.フロントエンド 美しい AI UI を構築するには、NextJS や Streamlit などのフレームワークが非常に役立ちます。また、Vercel はデプロイメントに役立ちます。 2.埋め込みと RAG ライブラリ Nomic、JinaAI、Cognito、LLMAware などの埋め込みモデルと RAG ライブラリは、開発者が正確な検索機能と RAG 機能を構築するのに役立ちます。 3.バックエンドとモデル アクセス バックエンド開発の場合、開発者は FastAPI、Langchain、Netflix Metaflow などのフレームワークを利用できます。モデル アクセスには、Ollama や Huggingface などのオプションが利用できます。 4.データと取得 データの保存と取得には、Postgres、Milvus、Weaviate、PGVector、FAISS などのいくつかのオプションが利用できます。 5.大規模言語モデル パフォーマンス ベンチマークに基づく、Llama、Mistral、Qwen、Phi、Gemma などのオープン ソース モデルは、GPT や Claude などの独自の LLM の優れた代替手段です。

EP146: オープンソース AI スタック - ByteByteGo ニュースレター

[hackernewsrobot]

Sharding Pgvector

https://pgdog.dev/blog/sharding-pgvector

[devnews]

Optimizing a RAG-Based Helpdesk Chatbot: Improving Accuracy with pgvector

Introduction In my previous post, I covered how I implemented pgvector in a RAG (Retrieval-Augmented Generation) system for a helpdesk chatbot. While the system performed well, accuracy issues arose: Some retrieved helpdesk articles weren’t fully relevant The chatbot sometimes misinterpreted the query context Long answers from the retrieved content confused GPT To fix these, I optimized the…

[vastperhaps]

Timescale Vector is PostgreSQL++ for AI applications | Timescale

[ericvanderburg]

Baffle Expands Data Protection to pgvector on PostgreSQL, Amplifying GenAI Security

http://securitytc.com/TGL2GX

[teknologinow]

Mengenal pgvector: Vector Database dari PostgreSQL

Apa itu pgvector? pgvector adalah ekstensi untuk PostgreSQL yang dirancang khusus untuk menyimpan dan melakukan pencarian terhadap data berbentuk vektor. Ekstensi ini memungkinkan Anda untuk menyimpan vektor dalam bentuk kolom pada tabel PostgreSQL dan melakukan pencarian berbasis kesamaan (similarity search) menggunakan berbagai metode. Hal ini membuat PostgreSQL mampu menangani data yang…

[vectordatabasecode]

Advanced Retrieval Techniques

Retrieval-Augmented Generation with Citations - Explore how augmentation with citations can significantly improve the depth and reliability of generated content.

Similarity Metrics for Vector Search - Understand different metrics that drive the effectiveness of vector searches, crucial for refining retrieval systems.

Local Agentic RAG with Langraph and Llama3 - Discover the integration of local datasets with advanced retrieval frameworks for enhanced performance.

Multimodal RAG with CLIP, Llama3, and Milvus - A deep dive into a multimodal approach, combining textual and visual data for rich content generation.

Practical Guides for Developers

A Beginner's Guide to Using Llama 3 with Ollama, Milvus, LangChain - Perfect for developers new to our frameworks, offering step-by-step guidance.

Getting Started with a Milvus Connection and Getting Started: Pgvector Guide for Developers Exploring Vector Databases - These guides are essential for setting up and beginning work with vector databases.

Educational Articles on Embedding Techniques and Applications

Sparse and Dense Embeddings - A look at different embedding types, offering insights into their use-cases and benefits.

Mastering BM25: A Deep Dive into the Algorithm and Application in Milvus - An in-depth exploration of BM25, a core algorithm for understanding document relevance.

Comparing SPLADE Sparse Vectors with BM25 - Comparative analysis that helps in selecting the right tool for specific retrieval tasks.

Training Your Own Text Embedding Model - Empower your projects by creating custom models tailored to your specific data needs.

Implementing and Optimizing RAG

Guide to Chunking Strategies for RAG and Experimenting with Different Chunking Strategies via LangChain - Both resources provide strategic insights into segmenting text for better retrieval outcomes.

Optimize RAG with Rerankers: The Role and Tradeoffs - Detailed discussion on the optimization of retrieval systems for balance between accuracy and performance.

[govindhtech]

Lightricks’ Videoleap Boost Search With Vector in Cloud SQL

Lightricks

With the help of a new AI-powered expansion, you can effectively express your next creative concept with captivating, dynamic pictures that perfectly capture your vision.

With the help of brand-new, AI-powered expansion, creatives who want to express and visualize thoughts can now accurately realize their ideas through eye-catching, dynamic visuals.

With three improved user workflows that let you realise your vision, you can take unparalleled control. Start your story with pre-edited content, or use custom AI technology to create a blank storyboard for complete creative control.

The difference is all in the delivery. Using the cast, style, storyline, and other assets you’ve created, export a concept video and complete slide deck. It’s now simpler than ever to put together everything you’ll need to propose your next project.

Reach your preferred aesthetic with ease. With just a few clicks, apply any style to your whole storyboard. Use automatic generations to replicate the look and feel of an image that you upload as a reference.

Savour sophisticated upscale features that let you export a high-resolution version of your creation. This implies that your images are reproduced with greater clarity and detail than they have ever been.

Light a creative fire inside of you

We believe the right gear will change your worldview regardless of your photography or videography experience. All of Lightricks products are designed to make it easier for people of all abilities to express themselves and experience the transformational power of creation.

Rethink the way you create

Lightricks studio develop user-friendly applications that enable content producers of all skill levels to turn their curiosity into awesome stuff. This translates to a lower learning curve and more user-friendly features that remove barriers to production and reinsert the joy of the creative process.

Change the way you appear online

Raise the bar on your inventiveness and give your work the consideration it merits. Make a connection with your audience and shape your community to best reflect your goal. Lightricks is available to assist you in navigating the dynamic landscape of the creative economy.

Videoleap

With the use of cutting-edge computer vision and artificial intelligence technology, Lightricks creates cutting-edge photo and video creation tools that let companies and content providers create scalable, highly engaging content. Using the pgvector extension in Cloud SQL for PostgreSQL, Lightricks was able to improve search capabilities and increase the number of retrieval rates by 40% for its robust video editor, Videoleap.

Google Cloud goal at Lightricks is to close the gap that exists between creation and imagination. With their video editing tool, Videoleap, both novices and experts can easily chop and combine movies no matter where they are.

With the use of AI algorithms, user-generated content (UGC) templates, and a simple editor, Google hope to make video editing accessible to everyone. Specifically, Videoleap’s template search feature is essential to their users’ ability to effectively browse this extensive and varied library of video templates. Google wanted a way to go towards a more dynamic search approach while still improving search. Google is realised that Cloud SQL for PostgreSQL was the best option for Videoleap when it revealed support for vector searches.

Seeking a solution that is superior to others

Google is already using Cloud SQL for PostgreSQL as their managed relational database before looking into vector database solutions. As a result, they were able to concentrate on improving their apps and spend less time managing the databases. However, in order to stay up with the expanding trends in video editing, Google required more support for Videoleap’s search features. Google’s goal was to enable consumers to select templates that rapidly matched their creative vision and to have more control over their browsing experience.

For a while, users had to utilize precise terms to retrieve relevant results from Google’s platform’s search feature, which depended on exact keyword matching based on predetermined annotations. This limited user flexibility. This method frequently fell short of capturing the wide range of user questions, which may contain intricate expressions or terminology that isn’t explicitly included in their annotations.

Since it would have taken a lot of time to resolve each of these differences on its own, Google chose to investigate the possibilities of vector search, which uses vector embeddings to extract pertinent data from databases. It didn’t take long for us to realise that adding this technique to Videoleap will improve search quality and speed while producing more context-aware results.

In order to meet Google objectives for search functionality, we looked at a few options. Pinecone and Vespa were Google’s first attempts, but their delayed integration and increased complexity made them unsatisfactory. Development overhead is greatly increased when a second vector database is introduced since it necessitates modifications to deployment procedures, pipelines for continuous integration, and local environments.

In addition, the learning curve for developers is higher than for pgvector, the well-known PostgreSQL extension for vector search, mostly because it requires them to comprehend a new data type. It can be difficult and error-prone to maintain data consistency between PostgreSQL and an external vector database, necessitating careful management of continuous data synchronization. It can also be difficult to take advantage of PostgreSQL’s transactional features for atomic updates across relational and vector data. The division of systems could potentially impede effective joins between vector data and other PostgreSQL tables.

Chroma Vector DB

Google also gave Chroma DB some thought, however it didn’t have any dependable hosting choices or deployment alternatives. Google therefore knew it was the appropriate decision when Cloud SQL for PostgreSQL introduced vector support via pgvector. It not only matched their requirements exactly, but it also worked seamlessly with the PostgreSQL architecture Google already had. For many use scenarios, its streamlined approach makes it a more dependable and efficient solution by lowering development overhead and minimizing the chance of data inconsistencies.

Using pgvector in conjunction with Cloud SQL to increase functionality

Google can effortlessly combine data, manage transactions, and enable semantic search by utilizing pgvector with Cloud SQL. Through the use of various indexing schemes for variables like speed and accuracy, Google may adjust and customize it to their specifications.

A microservices architecture is used in the application Google developed for Videoleap’s search function. It stores Cloud SQL with multiple embeddings of the templates and UGC template metadata to allow search while maintaining high availability and scalability. This method improves the modularity, flexibility, and scalability of the system while adhering to the concepts of microservices.Image credit to Google cloud

Faster results visualization with dynamic search features

Delivering relevant results for a wide range of queries was revolutionized by switching to a semantic search paradigm with vector embeddings. Rather than depending only on precise term matching, this modification was especially helpful for capturing the semantic intent of search queries. Because of this flexibility, Google can still get pertinent answers even if the query contains variations like synonyms, misspellings, or similar concepts. Instead of being constrained by the exact words used, a search for “labrador plays with frisbee” may now, for instance, yield a video of a golden retriever playing with a ball.

With the new method, Google observed a notable rise in export rates, suggesting that the modification to their search functionality undoubtedly added value for producers. Both the quantity of retrievals and the percentage of templates used from recovered results rose by 40% while using pgvector. The pgvector addition allowed us to query millions of embeddings with great precision and added support for the Hierarchical Navigable Small Worlds (HNSW) algorithm. 90% of Google’s inquiries (P90) had response times ranging from one to four seconds, but they now took less than 100 milliseconds. By enabling users to quickly find pertinent results, this unleashes the creative process and makes editing as gratifying as the original production.Image credit to Google cloud.  Hierarchical Navigable Small Worlds (HNSW) support enables high-accuracy querying of millions of embeddings. Response times (p90) plummeted from 1-4 seconds to under 100 milliseconds.

Using AI to bring search possibilities into greater focus

The addition of a visual content-based search capability is a key component of Google’s improved search capabilities. Neural networks can be used to generate vector embeddings with pgvector and Cloud SQL. As a result, they can now comprehend and match visual content with user intent with more accuracy and relevance, leading to more relevant search results.

Being aware of the emerging trend of AI-assisted editing, in which compositions are frequently based on textual cues, requires being up to date with the latest developments in video editing tools. For this reason, Google have a new Videoleap search function that uses AI suggestions to help content creators identify visual content that fits with more precise or subtle topics. One way to distinguish content would be to look for references to the Barbie movie vs images of generic Barbie doll models.

Read more on Govindhtech.com

[takachan]

PostgreSQLを生成AIの情報源として使える高速ベクトルデータベース化拡張「Pgvectorscale」がオープンソースで公開。Pgvectorをさらに高性能化

PostgreSQLのマネージドサービスなどを提供しているTimescaleは、PostgreSQLで高速なベクトルデータベース機能を実現する拡張機能「Pgvectorcale」をオープンソースとして公開したことを発表しました。 大規模言語... https://www.publickey1.jp/blog/24/postgresqlaipgvectorscalepgvector.html?utm_source=dlvr.it&utm_medium=tumblr Publickey

[honeycode]

Khoj is a Django project that runs PostgreSQL with pgvector for local vector embeddings. It contains good examples of solid, none-too-fancy ways to do local RAG.

[hackernewsrobot]

Pgvector Is Now Faster Than Pinecone at 75% Less Cost

https://www.timescale.com/blog/pgvector-is-now-as-fast-as-pinecone-at-75-less-cost/

[devnews]

Implementing a Vector Database in a RAG System for a Helpdesk Chatbot with pgvector

Introduction As AI-powered chatbots evolve, Retrieval-Augmented Generation (RAG) has become a crucial approach to improving their accuracy and contextual awareness. In this article, I’ll walk through how I implemented pgvector, a PostgreSQL extension for vector search, in a RAG-based helpdesk chatbot. One challenge when dealing with long helpdesk documents is ensuring that the chatbot retrieves…