MICE-P: end-to-end retrieval matching cross-encoder effectiveness with late-interaction throughput
_reachsumit · x · 2026-10-26
arXiv paper (2610.31932) introduces MICE-P, the first end-to-end retrieval pipeline for the MICE architecture, which indexes MICE's contextualized document token representations with PLAID. MICE had cross-encoder-like effectiveness via document-side pre-computation but lacked a practical storage scheme. In controlled comparisons training MICE, ColBERT and cross-encoders under matched conditions with the Ettin Family, MICE-P matches late-interaction retrieval throughput while retaining cross-encoder effectiveness, defining a new efficiency-effectiveness trade-off class. Code released.
More from Research
- HIT Shenzhen unveils STILL: content-aware intra-layer hybrid attention to linearize LLMs — jiqizhixin · 2026-10-26
- Bambi quietly gains nonlinear model support, with a blog post showing how — twiecki · 2026-10-26
- Fudan and Tencent unveil DraMaze, a multi-agent pipeline for diverse microdrama scripts — Siyang He · 2026-10-26
- Alibaba's MultiAgent-OPD trains hierarchical multi-agent systems end-to-end with on-policy distillation — TaoLiveAIGC · 2026-10-26
- LoCA: two-stage pipeline adds multi-modality to geospatial location encoders — Panagiotis Koromilas · 2026-10-26
- Skipping eval checks means you're likely fitting benchmark biases and noise — antoine_chaffin · 2026-10-26