Provable and Practical In-Context Policy Optimization for Self-Improvement

• URL: http://arxiv.org/abs/2603.01335v1
• Date: Mon, 02 Mar 2026 00:21:50 GMT
• Title: Provable and Practical In-Context Policy Optimization for Self-Improvement
• Authors: Tianrun Yu, Yuxiao Yang, Zhaoyang Wang, Kaixiang Zhao, Porter Jenkins, Xuchao Zhang, Chetan Bansal, Huaxiu Yao, Weitong Zhang,
• Abstract summary: We study test-time scaling, where a model improves its answer through multi-round self-reflection at inference.<n>We introduce In-Context Policy Optimization (ICPO), in which an agent optimize its response in context using self-assessed or externally observed rewards without modifying its parameters.<n>We propose Minimum-Entropy ICPO (ME-ICPO), a practical algorithm that iteratively uses its response and self-assessed reward to refine its response in-context at inference time.
• Score: 49.670847804409874
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: We study test-time scaling, where a model improves its answer through multi-round self-reflection at inference. We introduce In-Context Policy Optimization (ICPO), in which an agent optimizes its response in context using self-assessed or externally observed rewards without modifying its parameters. To explain this ICPO process, we theoretically show that with sufficient pretraining under a novel Fisher-weighted logit-matching objective, a single-layer linear self-attention model can provably imitate policy-optimization algorithm for linear bandits. Building on this theory, we propose Minimum-Entropy ICPO (ME-ICPO), a practical algorithm that iteratively uses its response and self-assessed reward to refine its response in-context at inference time. By selecting the responses and their rewards with minimum entropy, ME-ICPO ensures the robustness of the self-assessed rewards via majority voting. Across standard mathematical reasoning tasks, ME-ICPO attains competitive, top-tier performance while keeping inference costs affordable compared with other inference-time algorithms. Overall, ICPO provides a principled understanding of self-reflection in LLMs and yields practical benefits for test-time scaling for mathematical reasoning.

Related papers

• ICPO: Intrinsic Confidence-Driven Group Relative Preference Optimization for Efficient Reinforcement Learning [17.98065634130798]
  We propose the Intrinsic Confidence-Driven Group Relative Preference Optimization method (ICPO)<n>ICPO calculates a preference advantage score for each response by comparing the relative generation probabilities of multiple responses under the same input prompt.<n>We have discovered that the preference advantage score not only alleviates the issues of coarse-grained rewards and reward noise but also effectively curbs overconfident errors.
  arXiv  Detail & Related papers  (2025-11-26T03:10:15Z)
• Latent Chain-of-Thought for Visual Reasoning [53.541579327424046]
  Chain-of-thought (CoT) reasoning is critical for improving the interpretability and reliability of Large Vision-Language Models (LVLMs)<n>We reformulate reasoning in LVLMs as posterior inference and propose a scalable training algorithm based on amortized variational inference.<n>We empirically demonstrate that the proposed method enhances the state-of-the-art LVLMs on seven reasoning benchmarks.
  arXiv  Detail & Related papers  (2025-10-27T23:10:06Z)
• Learning to Refine: Self-Refinement of Parallel Reasoning in LLMs [102.48588475875749]
  We introduce Generative Self-Refinement (GSR), a novel parallel test-time scaling framework.<n>GSR generates a set of candidate responses in parallel and then performs self-refinement to synthesize a new superior solution.<n>We show that our method achieves state-of-the-art performance across five mathematical benchmarks.
  arXiv  Detail & Related papers  (2025-08-27T06:51:48Z)
• Truncated Proximal Policy Optimization [43.965892659920364]
  Truncated Proximal Policy Optimization (T-PPO) improves training efficiency by streamlining policy update and length-restricted response generation.<n>We propose Extended Generalized Advantage Estimation (EGAE) for advantage estimation derived from incomplete responses.<n>We demonstrate the effectiveness and efficacy of T-PPO on AIME 2024 with a 32B base model.
  arXiv  Detail & Related papers  (2025-06-18T01:21:38Z)
• Trajectory Bellman Residual Minimization: A Simple Value-Based Method for LLM Reasoning [55.33984461046492]
  Policy-based methods currently dominate reinforcement learning pipelines for large language model (LLM) reasoning.<n>We introduce Trajectory Bellman Residual Minimization (TBRM), an algorithm that naturally adapts this idea to LLMs.<n>We prove convergence to the near-optimal KL-regularized policy from arbitrary off-policy via an improved change-of-trajectory-measure analysis.
  arXiv  Detail & Related papers  (2025-05-21T09:41:53Z)
• Optimizing Anytime Reasoning via Budget Relative Policy Optimization [70.32755424260336]
  We present a novel framework, AnytimeReasoner, to optimize anytime reasoning performance.<n>We truncate the complete thinking process to fit within sampled token budgets from a prior distribution.<n>We then optimize the thinking and summary policies in a decoupled manner to maximize the cumulative reward.
  arXiv  Detail & Related papers  (2025-05-19T17:58:44Z)
• Provably Mitigating Overoptimization in RLHF: Your SFT Loss is Implicitly an Adversarial Regularizer [52.09480867526656]
  We identify the source of misalignment as a form of distributional shift and uncertainty in learning human preferences.<n>To mitigate overoptimization, we first propose a theoretical algorithm that chooses the best policy for an adversarially chosen reward model.<n>Using the equivalence between reward models and the corresponding optimal policy, the algorithm features a simple objective that combines a preference optimization loss and a supervised learning loss.
  arXiv  Detail & Related papers  (2024-05-26T05:38:50Z)
• $i$REPO: $i$mplicit Reward Pairwise Difference based Empirical Preference Optimization [12.266207199002604]
  Large Language Models (LLM) can sometimes produce outputs that deviate from human expectations. We propose a novel framework named $i$REPO, which utilizes implicit Reward pairwise difference regression for Empirical Preference Optimization. We show that $i$REPO effectively achieves self-alignment using soft-label, self-generated responses and the logit of empirical AI annotators.
  arXiv  Detail & Related papers  (2024-05-24T05:42:11Z)
• Overcoming Reward Overoptimization via Adversarial Policy Optimization with Lightweight Uncertainty Estimation [46.61909578101735]
  Adversarial Policy Optimization (AdvPO) is a novel solution to the pervasive issue of reward over-optimization in Reinforcement Learning from Human Feedback. In this paper, we introduce a lightweight way to quantify uncertainties in rewards, relying solely on the last layer embeddings of the reward model.
  arXiv  Detail & Related papers  (2024-03-08T09:20:12Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.