Ping Li Email & Phone Number

VecML (www.vecml.com) is committed to developing the cutting-edge infrastructures for building the next generation of LLM/AIGC applications, for accelerating accuracy, performance & cost-effectiveness with privacy protection of AIGC Infra. Currently, we are fast iterating our products. As the initial offering, we provide a free-trial SaaS service for the following: (1) neural search engine with privacy, including the standard vector search, search with sparse vectors, search with constraints (business filters), etc. Additional features such as search with neural net similarities and search with GPUs are also available upon request. (2) machine learning (ML) platform. Our ML platform is very fast, accurate, and easy-to-use. Both web UI and python interface are provided. To save cost, we place a limit on the training time. Please contact us if your company would like to have models with even higher accuracy. (3) Prompt generator, demoed as a chatbot. Users can upload private documents (such as PDF files), which are processed by our system to help generate the new prompt when a question arrive. The technologies behind the prompt generator include our neural search engine and ML platform. Users can also fairly easily build an LLM-empowered recommender system using our tools. This is just the beginning of our journey. We will frequently update our products to better serve the needs for developing the next generation of LLM/AIGC applications.

To build something cool in AI which can be useful for generations.

Sunnyvale, Ca, Us

https://pltrees.github.io/ 1. Advertising (ads) 2. Privacy: cohort-building, differential privacy (DP), federated learning 3. Embedding based retrieval (EBR)1 Privacy: Cohort, Differential privacy, federated learning(1) SIGIR 2023: ads privacy using cohort-building and CWS : https://arxiv.org/pdf/2304.13677.pdf (2) ICML 2023: compression for federated learning: https://proceedings.mlr.press/v202/li23o/li23o.pdf(3) NeurIPS 2023, differential privacy (DP) based on random projections and quantization:https://arxiv.org/pdf/2306.01751.pdf (4) differential privacy (DP) for sparse data, based on CWS and minhash:https://arxiv.org/pdf/2306.07674.pdf (5) Stochastic Controlled Averaging for Federated Learning with Communication Compression, https://arxiv.org/pdf/2308.08165.pdf2 Embedding-based retrieval, compressions: (6) Ads audience expansion, HNSW for sparse data, chi-square two-tower model: https://arxiv.org/pdf/2306.07607.pdf(7) KDD 2023: Fundamental data structure for embedding compression, privacy, and big models:https://pltrees.github.io/publication/KDD_2023_OPORP.pdfBig (large parameter) models for deep neural networks:(8) deep feature interactions using block-wise permutations: https://arxiv.org/pdf/2306.15881.pdf (9) Pb-Hash for large-parameter models (QR-hash is a special case):https://arxiv.org/pdf/2306.15944.pdfTalks in the public domain: 8/15 on boosting and trees: https://www.linkedin.com/feed/update/urn%3Ali%3Aactivity%3A6965026431579951104/ 8/30 on embedding-based retrieval: https://www.linkedin.com/feed/update/urn:li:activity:6972533869245865984/