TailRL works even with poor initial policy coverage, authors show in navigation tasks
YouJiacheng · x · 2026-09-08
stablegradients presents TailRL, addressing the objection that Best-of-k objectives fail when the initial policy is weak: in navigation tasks, the method still works even when the initial policy has poor coverage over good rollouts, by spending more training compute on a weighted Best-of-k objective. YouJiacheng notes even 0.01% coverage is valid. Part 6 of a thread.
Related event: CMU proposes TailRL: RL that optimizes tail rewards, not means(4 posts)→
More from Research
- Token Collapse: Study Finds Graph Soft Prompting Tokens in LLMs Fail to Encode Structure — Sapienza · 2026-10-27
- Mathathon 2.0 launches with Terence Tao's SAIR, $1.5M in compute for teams — _sathvikr · 2026-10-27
- The 'sweet lesson' is real: two kinds of ML problems need opposite strategies — mike64_t · 2026-10-27
- PixGR: Pixel-space generative decoding for fast feed-forward 3D reconstruction — kwangmoo_yi · 2026-10-27
- UniAction: unified action conditioning lets one world model handle mixed action types — ZeYanjie · 2026-10-27
- Infini-AI Lab open-sources Hermit, a framework for agent harness evolution and continual learning — BeidiChen · 2026-10-27