Linkup open-sources SPARSEUP, a sub-150M sparse embedding model, plus MICE reranking research
antoine_chaffin · x · 2026-10-20
Linkup Research hosts a Paris evening on information retrieval with three talks:
- SPARSEUP: Linkup's first open-source contribution—a sparse embedding model under 150M parameters, built on a simple recipe for training SPLADE-like encoders on modern backbones and data. In controlled comparisons with the same backbone and data it comes close to its dense counterpart (DenseOn) while trailing late-interaction models (LateOn). The talk also critically examines sparse retrieval's benchmark behavior and limits in large-scale production.
- MICE (Victor Morand, Sorbonne): a cross-encoder-derived architecture aiming to keep cross-encoder effectiveness while closing the efficiency gap with late-interaction approaches for reranking.
A useful reference for retrieval/RAG builders, plus lessons from shipping production search models.
More from Research
- Stanford and Databricks use AI agents to reproduce medical findings and test analytical robustness — ChengleiSi · 2026-10-23
- EMNLP paper: diffusion LMs internally encode latent denoising timestep representations — gsarti_ · 2026-10-23
- KU Leuven and VIB open two Group Leader professorships in AI for Biology, start 2027 — jpillowtime · 2026-10-23
- Subspace Tuning curbs LLM catastrophic forgetting by freezing principal singular directions — scottniekum · 2026-10-23
- Perceptic launches PercepticOS, an AI workspace for 'scientific superintelligence' — nathanbenaich · 2026-10-23
- EPFL's Bosselut to talk long context as test-time learning at ELLIS seminar — andre_t_martins · 2026-10-23