4h ago

Researchers Replace K-Means With Sparse Autoencoders For Multi-Vector Retrieval

——0——
Original post
OPOmar Khattab#160Omar Khattab@LATEINTERACTION

Late-interaction sparse retrieval? 😁 With neuron-level inverted indexing, on top of unsupervised sparse autoencoders. Works much better than directly training sparse retrievers. Lots of cool ideas developed & composed in here. Thanks for the insights @Veritas2026 @yifeiwang77!

5:35 PM · May 29, 2026 View on X
REPLYOmar Khattab#160Omar Khattab@LATEINTERACTION

i guess saying “works much better than directly training sparse retrievers” leaves out that this is multi-vector vs. single-vector. still trying to wrap my head around the choices here!

also, why is this tweet attracting so many bots lol :/

Omar KhattabOmar Khattab@lateinteraction

Late-interaction sparse retrieval? 😁 With neuron-level inverted indexing, on top of unsupervised sparse autoencoders. Works much better than directly training sparse retrievers. Lots of cool ideas developed & composed in here. Thanks for the insights @Veritas2026 @yifeiwang77!

12:35 AM · May 30, 2026 · 6.2K Views
1:35 AM · May 30, 2026 · 1.2K Views