AURA Teaches Retrieval Models When to Stop Reasoning, Cutting Latency from 4.6s to 0.76s
_reachsumit · x · 2026-10-16
- A paper by Yingtao Luo et al. introduces AURA, accepted to EMNLP 2026 Findings, addressing how extra reasoning can inject noisy abstractions that hurt retrieval specificity.
- AURA directly supervises intermediate reasoning states by how well they separate relevant documents from hard distractors, and trains a lightweight utility estimator over prefix-conditioned embeddings to adaptively stop reasoning at inference time.
- On the reasoning-intensive BRIGHT benchmark, AURA-4B reaches 34.7 average nDCG@10 using only 2.8 reasoning steps and 96 generated tokens per query.
- Under matched deployment conditions, adaptive stopping cuts latency from 4.6s to 0.76s, with MS MARCO results showing minimal degradation on conventional retrieval.
More from Research
- Beam LoRA fine-tuning benchmark: 500 steps, 16M tokens, 16 minutes on one 8x B300 node — QGallouedec · 2026-10-16
- Transforming quadruped robot runs on real hardware: per-pose RL policies in Isaac switched on morph — _Stocko_ · 2026-10-16
- KinoGen synthesizes robot loco-manipulation reference motions directly via kinematically grounded video generation — ChongZzZhang · 2026-10-16
- Science Robotics paper introduces F³, predictive event-based vision beyond optical flow — CSProfKGD · 2026-10-16
- VideoGen-Agent: An 8B Multimodal Agent Turns Video Generation Into a Tool-Use Loop — jiqizhixin · 2026-10-16
- Comprehension Audits: New Paper Proposes a Gate for AI-Written Code at Frontier Labs — ronbodkin · 2026-10-16