Which Rollout Taught It That? BehaviorTrace and the Limits of Training-Data Attribution in Online RL
Paper recorded by Signals 4 on 2026-10-07 in cs.LG. Abstract reproduced from arXiv; link to the original below.
Published 2026-10-07 on arXiv · recorded by Signals 4 on 2026-10-08
Category: cs.LG · 机器学习 · first seen 2026-10-08
Abstract
When reinforcement learning teaches a language model a new behavior, can we find the training rollouts that taught it? And when an attribution method says it can, how do we know the answer is real? We study both questions on online RL fine-tuning with GRPO, using a planted behavior with a known cause. We release BehaviorTrace, an open evaluation harness that combines full-gradient sketching, the p
Read on arXiv →
Cite this page: Which Rollout Taught It That? BehaviorTrace and the Limits of Training-Data Attribution in Online RL: the #355 most recent of 762 cs.LG papers we have recorded (as of 2026-10-07). Source: Signals 4 (Signals API) — https://data.jiangzhang.ca/signals4/t/papers/which-rollout-taught-it-that-behaviortrace-and-the-limits-of-training-data-attri.html
Free to quote with attribution to “Signals 4 (Signals API)”. Machine-readable:
papers.json
Get 4 AI signals a day by email — free.
Get 4 AI signals a day by email — free