Research talk: Approximate nearest neighbor search systems at scale
3 381
15.4
Microsoft Research334 тыс
Следующее
25.01.22 – 1732:08Популярные
11.11.23 – 5731:03:57
Опубликовано 25 января 2022, 1:37
Speaker: Harsha Simhadri, Principal Researcher, Microsoft Research India
Building deep learning-based search and recommendation systems at internet scale requires a complete redesign of the search index. Key to this redesign is a fast, accurate, and cost-efficient indexing system for approximate nearest neighbor search. In this talk, we’ll present our recent advances in this space, including the DiskANN and FreshDiskANN systems and the underlying algorithms. These algorithms present an order-of-magnitude improvement in scale and cost-of-operation over the state of the art and are a first of their kind at effectively using solid-state drives (SSDs) to serve at interactive (milliseconds) latencies. In addition, they provide faster in-memory search than other graph indices, like HNSW, and support real-time concurrent insertions and deletions to SSD-resident indices without losing recall. We’ll provide an overview their applicability to various product scenarios and highlight directions for further research.
Learn more about the 2021 Microsoft Research Summit: Aka.ms/researchsummit
Building deep learning-based search and recommendation systems at internet scale requires a complete redesign of the search index. Key to this redesign is a fast, accurate, and cost-efficient indexing system for approximate nearest neighbor search. In this talk, we’ll present our recent advances in this space, including the DiskANN and FreshDiskANN systems and the underlying algorithms. These algorithms present an order-of-magnitude improvement in scale and cost-of-operation over the state of the art and are a first of their kind at effectively using solid-state drives (SSDs) to serve at interactive (milliseconds) latencies. In addition, they provide faster in-memory search than other graph indices, like HNSW, and support real-time concurrent insertions and deletions to SSD-resident indices without losing recall. We’ll provide an overview their applicability to various product scenarios and highlight directions for further research.
Learn more about the 2021 Microsoft Research Summit: Aka.ms/researchsummit
Свежие видео
3 дня – 31 1919:52
4 дня – 874 5670:59
6 дней – 7 2752:08
6 дней – 6 5971:01
Случайные видео
327 дней – 92 61920:29
354 дня – 1 003 9329:41
29.05.22 – 1 8463:33
16.07.21 – 446 5468:43
18.03.14 – 1 2251:27
2 дня – 977 54718:29
3 дня – 5820:07
4 дня – 5 01458:36
6 дней – 120 77426:23
10 дней – 1 1378:34
10 дней – 1 0700:58
