Using Tensor and Rank Profile Math to Combine Dense and Sparse Signals скачать в хорошем качестве

Using Tensor and Rank Profile Math to Combine Dense and Sparse Signals

Welcome to the 25th London Information Retrieval & AI Meetup, a free evening event aimed at Information Retrieval and AI enthusiasts and professionals who are curious to explore and discuss the latest trends in the field. The first talk is from Radu Gheorghe – Software Engineer @ Vespa. Title: "Using Tensor and Rank Profile Math to Combine Dense and Sparse Signals" Abstract: "We’ll take a top-down approach to retrieval: which signals matter most (dense/sparse vectors, lexical…) and we’ll look at how to combine them in various ways (including normalising) to get good quality. In Vespa, we’d recommend modeling relevance with tensor math wherever possible: it’s much faster and potentially easier to maintain. But we’ll also look at combining higher-level signals (e.g., from document fields) into rank profiles. Rank profiles are Vespa’s relevance language, exposing features like inheritance and 2-stage re-ranking. To exemplify, we’ll show open-source sample applications that deal with both dense vectors (ColPali) and sparse (LLM-generated user preferences)." If you are willing to attend the next London Information Retrieval & AI Meetup, don't forget to join our group: https://bit.ly/2IjSBeX We are also accepting talks for the next meetups. If you have any talk you would like to propose, feel free to send us an abstract at talk@sease.io. ********************************** Hosted by Sease: https://sease.io