Whether an adaptive , knowledge-aware curriculum that mirrors this behaviour improves RLVR reasoning is asked, and—since the question also bears on whether RLVR creates new ability or merely sharpens existing ability—the authors study both on Countdown.
This work proposes a self-evolving curriculum that, each round, evaluates the current checkpoint, seeds QbQ from the problems it can mostly get right, and trains on the resulting variants; under an identical data budget, this breaks the ceiling and lifts pass@1 to 16.5\% with no sign of saturation after 20 rounds.
Longtian Bao, Jianyou Wang, Yang Zhang et al.· 0 citations
A quantitative account of the pretraining-to-RL interface and a controlled testbed for studying the science of reasoning across the full pretraining-to-post-training pipeline is provided.
Jingyan Shen, Ang Li, Salman Rahman et al.· 0 citations
Learning from Hindsight is presented, which brings hindsight relabeling to RL post-training of VLAs by scoring failed rollouts against the tasks they actually achieved, and achieves 5 times improvement in sample efficiency, and outperforms a dense progress-reward baseline on out-of-distribution LIBERO-PRO tasks.
Iris Xu, Sunshine Jiang, John Marangola et al.· arXiv.org· 0 citations
It is shown that the transfer relationship between different difficulty levels characterizes the optimization dynamics induced by curriculum learning, which in turn explains the effectiveness of different curriculum schedules, and formalize this relationship as Relative Transfer, a principled measure of cross-difficult...
Evidence is found for two mechanisms behind the learning gains: students shift time away from drafting text and toward reading and searching for information, and they report greater learning enjoyment.
Instruction-Conditioned Exploration (ICE), which appends one of a small fixed set of instructions to task prompts during training, using the same set for every problem, increasing the coverage of behaviours attempted, improves Qwen3-1.7B held-out pass@1 performance at 4K response length on mathematical reasoning tasks.
Jim Dilkes, V. Yazdanpanah, Sebastian Stein· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
