Reproduce “MICE to Meet You: From Re-ranking Only to End-to-End Retrieval” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Victor Morand, Yuxuan Zong, Mathias Vast, Josiane Mothe, Benjamin Piwowarski
MICE-P extends MICE into a retrieval pipeline by storing compressed document representations, retrieving top-100 candidates with SPLADE, and scoring them with MICE’s interaction layers. A 128-dimensional bottleneck enables storage in PLAID’s compressed format, but PLAID does not retrieve candidates. On 13 decontaminated BEIR datasets, MICE-P-130m scores 57.9 average nDCG@10 versus 58.8 for Ettin-reranker-150m, with additional indexing costs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.