CPR-Dataset: retrievers fail at cross-domain discovery, training gives +70% on BRIGHT
coallaoh · x · 2026-10-28
Researchers introduced CPR-Dataset (Cross-Pollination Retrieval) to test whether AI can surface idea connections across scientific fields. Key points:
- Existing retrievers perform poorly on this task, with the best at only 0.26 Recall@10; unlike SciDocs/LitSearch where queries and relevant papers stay in a narrow domain, CPR pairs span health, life, physical and social sciences.
- Training on CPR transfers well: up to +33% on AnaloBench and +70% on BRIGHT.
The result shows cross-domain analogical retrieval is a clear weakness of current models — and one that targeted training can substantially fix.
More from Research
- Effective learning rate explains peak-then-decay schedules as bias-variance optimum — burny_tech · 2026-10-28
- Formalizing the classification of finite simple groups: a draw between machines and humans — burny_tech · 2026-10-28
- Paper: "bottleneck pressure" metric shows where AI-infeasible work gains value as AI speeds up the rest — soumitrashukla9 · 2026-10-28
- New IEEE IT paper: geometric Jensen-Shannon divergence between Gaussian measures on Hilbert space — FrnkNlsn · 2026-10-28
- One-page Q-learning notes with math, code and a playground for RL beginners — East-Muffin-6472 · 2026-10-28
- Hugging Face robot duck walks in simulated snow with 290,000 physics-simulated flakes — lukas_m_ziegler · 2026-10-28