Paper Detail
Aozhe Wang, Zhengxi Lu, Jianze Wang, Shangke Lv, Ying Liu, Weiming Lu, Jun Xiao, Yueting Zhuang, Hua Yang, Qianglong Chen, Yongliang Shen
Recent prominent post-training methods, such as Reinforcement Learning (RL) and On-Policy Self-Distillation (OPSD), have driven rapid progress in mathematical reasoning for large language models, yet their reliance on ground-truth labels precludes test-time training (TTT). Replacing ground truth with majority-vote pseudo-labels is a natural alternative, yet it is fragile: an incorrect vote corrupts the teacher and misleads every token. We observe that this failure mode is asymmetric: rollouts that disagree with the pseudo-label are typically wrong regardless of whether the vote itself is correct. Building on this observation, we propose Test-Time Policy Optimization (TTPO), an asymmetric objective that distills agreeing rollouts via OPSD and penalizes disagreeing rollouts with Grouped RL. Token-level selection further refines both branches: distillation down-weights already-converged positions, while RL penalizes only confident errors. Both updates remain well-grounded even under frequent pseudo-label errors, and majority-vote routing yields tighter self-supervision as the model improves. Without any labels, TTPO matches label-supervised OPSD on five competition-level benchmarks, raises Qwen3-1.7B from 38.0% to 45.2% in TTT, yields +25.2% to +36.4% without thinking, and shows strong cross-task generalization.
No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.
No ranking explanation is available yet.
No tags.
@article{wang2026ttpo,
title = {TTPO: Test-Time Policy Optimization},
author = {Aozhe Wang and Zhengxi Lu and Jianze Wang and Shangke Lv and Ying Liu and Weiming Lu and Jun Xiao and Yueting Zhuang and Hua Yang and Qianglong Chen and Yongliang Shen},
year = {2026},
abstract = {Recent prominent post-training methods, such as Reinforcement Learning (RL) and On-Policy Self-Distillation (OPSD), have driven rapid progress in mathematical reasoning for large language models, yet their reliance on ground-truth labels precludes test-time training (TTT). Replacing ground truth with majority-vote pseudo-labels is a natural alternative, yet it is fragile: an incorrect vote corrupts the teacher and misleads every token. We observe that this failure mode is asymmetric: rollouts th},
url = {https://arxiv.org/abs/2608.27448},
keywords = {cs.CL, Reinforcement Learning, On-Policy Self-Distillation, test-time training, majority-vote pseudo-labels, Test-Time Policy Optimization, Grouped RL, token-level selection, code available, huggingface daily, Computer science, Reinforcement learning, Economics, Selection (genetic algorithm), Work (physics)},
eprint = {2608.27448},
archiveprefix = {arXiv},
}
{}