Report
MICE expands beyond re-ranking into end-to-end retrieval
A paper author says the architecture turns theoretical pre-computation gains into an operational retrieval system.
TLDR
A paper author says their EMNLP 2026 work showed MICE pushing the cross-encoder Pareto frontier, and that they have since extended it into an end-to-end retrieval architecture. Another poster welcomed open releases for making comparisons more apples-to-apples than before, while noting those comparisons still aren't perfect.
Combined views
1.4K
3 Sources, first seen ago
