Vector Podcast

Topics:00:00 Kick-off by Judy Zhu01:33 Introduction by Dmitry Kan and his bio!03:03 Daniel’s background04:46 “Science is the difference between instinct and strategy” 07:41 Search as a personal learning experience11:53 Why do we need Machine Learning in Search, or can we use manually curated features?16:47 Swimming up-stream from relevancy: query / content understanding and where to start?23:49 Rule-based vs Machine Learning approaches to Query Understanding: Pareto principle29:05 How content understanding can significantly improve your search engine experience32:02 Available datasets, tools and algorithms to train models for content understanding38:20 Daniel’s take on the role of vector search in modern search engine design as the path to language of users45:17 Mystical question of WHY: what drives Daniel in the search space today49:50 Announcements from Daniel51:15 Questions from the audienceShow notes:[What is Content Understanding?. Content understanding is the foundation… | by Daniel Tunkelang | Content Understanding | Medium](https://medium.com/content-understanding/what-is-content-understanding-4da20e925974)[Query Understanding: An Introduction | by Daniel Tunkelang | Query Understanding](https://queryunderstanding.com/introduction-c98740502103)Science as Strategy [YouTube](https://www.youtube.com/watch?v=dftt6Yqgnuw)Search Fundamentals course - https://corise.com/course/search-fundamentalsSearch with ML course - https://corise.com/course/search-with-machine-learningBooks:Faceted Search, by Daniel Tunkelang: https://www.amazon.com/Synthesis-Lectures-Information-Concepts-Retrieval/dp/1598299999Modern Information Retrieval: The Concepts and Technology Behind Search, by Ricardo Baeza-Yates: https://www.amazon.com/Modern-Information-Retrieval-Concepts-Technology/dp/0321416910/ref=sr11?qid=1653144684&refinements=p_27%3ARicardo+Baeza-Yates&s=books&sr=1-1Introduction to Information Retrieval, by Chris Manning: https://www.amazon.com/Introduction-Information-Retrieval-Christopher-Manning/dp/0521865719/ref=sr1fkmr0_1?crid=2GIR19OTZ8QFJ&keywords=chris+manning+information+retrieval&qid=1653144967&s=books&sprefix=chris+manning+information+retrieval%2Cstripbooks-intl-ship%2C141&sr=1-1-fkmr0Query Understanding for Search Engines, by Yi Chang and Hongbo Deng: https://www.amazon.com/Understanding-Search-Engines-Information-Retrieval/dp/3030583333

Topics:00:00 Intro01:03 Yusuf’s background03:00 Multimodal search in tech and humans08:53 CLIP: discovering hidden semantics13:02 Where to start to apply metric learning in practice. AutoEncoder architecture included!19:00 Unpacking it further: what is metric learning and the difference with deep metric learning?28:50 How Deep Learning allowed us to transition from pixels to meaning in the images32:05 Increasing efficiency: vector compression and quantization aspects34:25 Yusuf gives a practical use-case with Conversational AI of where metric learning can prove to be useful. And tools!40:59 A few words on how the podcast is made :) Yusuf’s explanation of how Gmail smart reply feature works internally51:19 Metric learning helps us learn the best vector representation for the given task52:16 Metric learning shines in data scarce regimes. Positive impact on the planet58:30 Yusuf’s motivation to work in the space of vector search, Qdrant, deep learning and metric learning — the question of Why1:05:02 Announcements from Yusuf- Join discussions at Discord: https://discord.qdrant.tech - Yusuf's Medium: https://medium.com/@yusufsarigoz and LinkedIn: https://www.linkedin.com/in/yusufsarigoz/ - GSOC 2022: TensorFlow Similarity - project led by Yusuf: https://docs.google.com/document/d/1fLDLwIhnwDUz3uUV8RyUZiOlmTN9Uzy5ZuvI8iDDFf8/edit#heading=h.zftd93u5hfnp - Dmitry's Twitter: https://twitter.com/DmitryKanFull Show Notes: https://www.youtube.com/watch?v=AU0O_6-EY6s

Topics:00:00 Introduction01:21 Jo Kristian’s background in Search / Recommendations since 2001 in Fast Search & Transfer (FAST)03:16 Nice words about Trondheim04:37 Role of NTNU in supplying search talent and having roots in FAST 05:33 History of Vespa from keyword search09:00 Architecture of Vespa and programming language choice: C++ (content layer), Java (HTTP requests and search plugins) and Python (pyvespa)13:45 How Python API enables evaluation of the latest ML models with Vespa and ONNX support17:04 Tensor data structure in Vespa and its use cases22:23 Multi-stage ranking pipeline use cases with Vespa24:37 Optimizing your ranker for top 1. Bonus: cool search course mentioned!30:18 Fascination of Query Understanding, ways to implement and its role in search UX33:34 You need to have investment to get great results in search35:30 Game-changing vector search in Vespa and impact of MS Marco Passage Ranking38:44 User aspect of vector search algorithms43:19 Approximate vs exact nearest neighbor search tradeoffs47:58 Misconceptions in neural search52:06 Ranking competitions, idea generation and BERT bi-encoder dream56:19 Helping wider community through improving search over CORD-19 dataset58:13 Multimodal search is where vector search shines1:01:14 Power of building fully-fledged demos1:04:47 How to combine vector search with sparse search: Reciprocal Rank Fusion1:10:37 The philosophical WHY question: Jo Kristian’s drive in the search field1:21:43 Announcement on the coming features from Vespa- Jo Kristian’s Twitter: https://twitter.com/jobergum- Dmitry’s Twitter: https://twitter.com/DmitryKanFor the Show Notes check: https://www.youtube.com/watch?v=UxEdoXtA9oM

Update: ZIR.AI has relaunched as Vectara: https://vectara.com/Topics:00:00 Intro00:54 Amin’s background at Google Research and affinity to NLP and vector search field05:28 Main focus areas of ZIR.AI in neural search07:26 Does the company offer neural network training to clients? Other support provided with ranking and document format conversions08:51 Usage of open source vs developing own tech10:17 The core of ZIR.AI product14:36 API support, communication protocols and P95/P99 SLAs, dedicated pools of encoders17:13 Speeding up single node / single customer throughput and challenge of productionizing off the shelf models, like BERT23:01 Distilling transformer models and why it can be out of reach of smaller companies25:07 Techniques for data augmentation from Amin’s and Dmitry’s practice (key search team: margin loss)30:03 Vector search algorithms used in ZIR.AI and the need for boolean logic in company’s client base33:51 Dynamics of open source in vector search space and cloud players: Google, Amazon, Microsoft36:03 Implementing a multilingual search with BM25 vs neural search and impact on business38:56 Is vector search a hype similar to big data few years ago? Prediction for vector search algorithms influence relations databases43:09 Is there a need to combine BM25 with neural search? Ideas from Amin and features offered in ZIR.AI product51:31 Increasing the robustness of search — or simply making it to work55:10 How will Search Engineer profession change with neural search in the game?Get a $100 discount (first month free) for a 50mb plan, using the code VectorPodcast (no lock-in, you can cancel any time): https://zir-ai.com/signup/user

Topics:00:00 Introduction01:04 Yury’s background in laser physics, computer vision and startups05:14 How Yury entered the field of nearest neighbor search and his impression of it09:03 “Not all Small Worlds are Navigable”10:10 Gentle introduction into the theory of Small World Navigable Graphs and related concepts13:55 Further clarification on the input constraints for the NN search algorithm design15:03 What did not work in NSW algorithm and how did Yury set up to invent new algorithm called HNSW24:06 Collaboration with Leo Boytsov on integrating HNSW in nmslib26:01 Differences between HNSW and NSW27:55 Does algorithm always converge?31:56 How FAISS’s implementation is different from the original HNSW33:13 Could Yury predict that his algorithm would be implemented in so many frameworks and vector databases in languages like Go and Rust?36:51 How our perception of high-dimensional spaces change compared to 3D?38:30 ANN Benchmarks41:33 Feeling proud of the invention and publication process during 2,5 years!48:10 Yury’s effort to maintain HNSW and its GitHub community and the algorithm’s design principles53:29 Dmitry’s ANN algorithm KANNDI, which uses HNSW as a building block1:02:16 Java / Python Virtual Machines, profiling and benchmarking. “Your analysis of performance contradicts the profiler”1:05:36 What are Yury’s hopes and goals for HNSW and role of symbolic filtering in ANN in general1:13:05 The future of ANN field: search inside a neural network, graph ANN1:15:14 Multistage ranking with graph based nearest neighbor search1:18:18 Do we have the “best” ANN algorithm? How ANN algorithms influence each other1:21:27 Yury’s plans on publishing his ideas1:23:42 The intriguing question of WhyShow notes:- HNSW library: https://github.com/nmslib/hnswlib/- HNSW paper Malkov, Y. A., & Yashunin, D. A. (2018). Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs. TPAMI, 42(4), 824-836. (arxiv:1603.09320)- NSW paper Malkov, Y., Ponomarenko, A., Logvinov, A., & Krylov, V. (2014). Approximate nearest neighbor algorithm based on navigable small world graphs. Information Systems, 45, 61-68.- Yury Lifshits’s paper: https://yury.name/papers/lifshits2009combinatorial.pdf- Sergey Brin’s work in nearest neighbour search: GNAT - Geometric Near-neighbour Access Tree: [CiteSeerX — Near neighbor search in large metric spaces](http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.173.8156)- Podcast with Leo Boytsov: https://rare-technologies.com/rrp-4-leo-boytsov-knn-search/- Million-Scale ANN Benchmarks: http://ann-benchmarks.com/- Billion Scale ANN Benchmarks: https://github.com/harsha-simhadri/big-ann-benchmarks- FALCONN algorithm: https://github.com/falconn-lib/falconn- Mentioned navigable small world papers: Kleinberg, J. M. (2000). Navigation in a small world. Nature, 406(6798), 845-845.; Boguna, M., Krioukov, D., & Claffy, K. C. (2009). Navigability of complex networks. Nature Physics, 5(1), 74-80.

Topics:00:00 Intro00:42 Joan's background01:46 What attracted Joan's attention in Jina as a company and product?04:39 Main area of focus for Joan in the product05:46 How Open Source model works for Jina?08:38 Deeper dive into Jina.AI as a product and technology stack11:57 Does Jina fit the use cases of smaller / mid-size players with smaller amount of data?13:45 KNN/ANN algorithms available in Jina16:05 BigANN competition and BuddyPQ, increasing 12% in recall over FAISS17:07 Does Jina support customers in model training? Finetuner20:46 How does Jina framework compare to Vector Databases?26:46 Jina's investment in user-friendly APIs31:04 Applications of Jina beyond search engines, like question answering systems33:20 How to bring bits of neural search into traditional keyword retrieval? Connection to model interpretability41:14 Does Jina allow going multimodal, including images / audio etc?46:03 The magical question of Why55:20 Product announcement from JoanOrder your Jina swag https://docs.google.com/forms/d/e/1FAIpQLSedYVfqiwvdzWPX-blCpVu-tQoiFiUJQz2QnIHU1ggy1oyg/ Use this promo code: vectorPodcastxJinaAIShow notes:- Jina.AI: https://jina.ai/- HNSW + PostgreSQL Indexer: [GitHub - jina-ai/executor-hnsw-postgres: A production-ready, scalable Indexer for the Jina neural search framework, based on HNSW and PSQL](https://github.com/jina-ai/executor-h...)- pqlite: [GitHub - jina-ai/pqlite: A fast embedded library for Approximate Nearest Neighbor Search integrated with the Jina ecosystem](https://github.com/jina-ai/pqlite)- BuddyPQ: [Billion-Scale Vector Search: Team Sisu and BuddyPQ | by Dmitry Kan | Big-ANN-Benchmarks | Nov, 2021 | Medium](https://medium.com/big-ann-benchmarks...)- PaddlePaddle: [GitHub - PaddlePaddle/Paddle: PArallel Distributed Deep LEarning: Machine Learning Framework from Industrial Practice （『飞桨』核心框架，深度学习&机器学习高性能单机、分布式训练和跨平台部署）](https://github.com/PaddlePaddle/Paddle)- Jina Finetuner: [Finetuner 0.3.1 documentation](https://finetuner.jina.ai/)- [Not All Vector Databases Are Made Equal | by Dmitry Kan | Towards Data Science](https://towardsdatascience.com/milvus...)- Fluent interface (method chaining): [Fluent interfaces in Python | Florian Einfalt – Developer](https://florianeinfalt.de/posts/fluen...)- Sujit Pal’s blog: [Salmon Run](http://sujitpal.blogspot.com/)- ByT5: Towards a token-free future with pre-trained byte-to-byte models https://arxiv.org/abs/2105.13626Special thanks to Saurabh Rai for the Podcast Thumbnail: https://twitter.com/srbhr_ https://www.linkedin.com/in/srbh077/

Show notes:- The ML Test Score: A Rubric for ML Production Readiness and Technical Debt Reduction https://research.google/pubs/pub46555/- IEEE MLOps Standard for Ethical AI https://docs.google.com/document/d/1x...- Qdrant: https://qdrant.tech/- Elixir connector for Qdrant by Tom: https://github.com/tlack/exqdr- Other 6 vector databases: https://towardsdatascience.com/milvus...- ByT5: Towards a token-free future with pre-trained byte-to-byte models https://arxiv.org/abs/2105.13626- Tantivy: https://github.com/quickwit-inc/tantivy- Papers with code: https://paperswithcode.com/

Show notes:- On the Measure of Intelligence by François Chollet - Part 1: Foundations (Paper Explained) [YouTube](https://www.youtube.com/watch?v=3_qGr...)- [2108.07258 On the Opportunities and Risks of Foundation Models](https://arxiv.org/abs/2108.07258)- [2005.11401 Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401)- Negative Data Augmentation: https://arxiv.org/abs/2102.05113- Beyond Accuracy: Behavioral Testing of NLP models with CheckList: [2005.04118 Beyond Accuracy: Behavioral Testing of NLP models with CheckList](https://arxiv.org/abs/2005.04118)- Symbolic AI vs Deep Learning battle https://www.technologyreview.com/2020...- Dense Passage Retrieval for Open-Domain Question Answering https://arxiv.org/abs/2004.04906- Data Augmentation Can Improve Robustness https://arxiv.org/abs/2111.05328- Contrastive Loss Explained. Contrastive loss has been used recently… | by Brian Williams | Towards Data Science https://towardsdatascience.com/contra...- Keras Code examples https://keras.io/examples/- https://you.com/ -- new web search engine by Richard Socher- The Book of Why: The New Science of Cause and Effect: Pearl, Judea, Mackenzie, Dana: 9780465097609: Amazon.com: Books https://www.amazon.com/Book-Why-Scien...- Chelsea Finn: https://twitter.com/chelseabfinn- Jeff Clune: https://twitter.com/jeffclune- Michael Bronstein (Geometric Deep Learning): https://twitter.com/mmbronstein https://arxiv.org/abs/2104.13478- Connor's Twitter: https://twitter.com/CShorten30- Dmitry's Twitter: https://twitter.com/DmitryKan

Order your Milvus t-shirt / hoodie! https://milvus.typeform.com/to/IrnLAgui Thanks Filip for arranging.Show notes: - Milvus DB: https://milvus.io/ - Not All Vector Databases Are Made Equal: https://towardsdatascience.com/milvus... - Milvus talk at Haystack: https://www.youtube.com/watch?v=MLSMs... - BEIR: A Heterogenous Benchmark for Zero-shot Evaluation of Information Retrieval Models https://arxiv.org/abs/2104.08663 - End-to-End Environmental Sound Classification using a 1D Convolutional Neural Network: https://arxiv.org/abs/1904.08990 - What BERT is not: Lessons from a new suite of psycholinguistic diagnostics for language models https://arxiv.org/abs/1907.13528 - NVIDIA Triton Inference Server: https://developer.nvidia.com/nvidia-t... - Towhee -- ML / Embedding pipeline making steps before Milvus easier: https://github.com/towhee-io/towhee - Being at the leading edge: http://paulgraham.com/startupideas.html

1. Layering problem: www.edge.org/conversation/sean_…-layers-of-reality2. Podcast with Etienne Dilocker (SeMI Technologies Co-Founder & CTO): www.youtube.com/watch?v=6lkanzOqhDs3. SOC2: linfordco.com/blog/soc-1-vs-soc-2-audit-reports/4. Dmitry's post on 7 Vector Databases: towardsdatascience.com/milvus-pineco…-9c65a3bd06965. Billion-Scale ANN Challenge: big-ann-benchmarks.com/index.html6. Weaviate Introduction: www.semi.technology/developers/weaviate/current/ Newsletter: www.semi.technology/newsletter/7. Use case: Scalable Knowledge Graph Search for 60+ million academic papers with Weaviate: medium.com/keenious/knowledge-…aviate-7964657ec9118. Bob's Twitter: twitter.com/bobvanluijt9. Dmitry's Twitter: twitter.com/DmitryKan10. Dmitry's tech blog: dmitry-kan.medium.com/