ECHO: Entropy-Confidence Hybrid Optimization for Test-Time Reinforcement Learning

Test-time reinforcement learning generates multiple candidate answers via repeated rollouts and performs online updates using pseudo-labels constructed by majority voting. To reduce overhead and improve exploration, prior work introduces tree-structured rollouts, which share reasoning prefixes and branch at key nodes to improve sampling efficiency. However, this paradigm still faces two challenges: (1) high-entropy branching can trigger rollout collapse, where the branching budget concentrates on a few trajectories with consecutive high-entropy segments, rapidly reducing the number of effective branches; (2) early pseudo-labels are noisy and biased, which can induce self-reinforcing overfitting, causing the policy to sharpen prematurely and suppress exploration. To address these issues, we propose Entropy–Confidence Hybrid Group Relative Policy Optimization (ECHO). During rollout, ECHO jointly leverages local entropy and group-level confidence to adaptively control branch width, and further introduces online confidence-based pruning to terminate persistently low-confidence branches, avoiding high-entropy traps and mitigating collapse. During policy updates, ECHO employs confidence-adaptive clipping and an entropy–confidence hybrid advantage shaping approach to enhance training robustness and mitigate early-stage bias. Experiments demonstrate that ECHO achieves consistent gains on multiple mathematical and visual reasoning benchmarks, and generalizes more effectively under a limited rollout budget.

When solving reasoning problems, large language models can improve by generating multiple candidate answers and learning from the most common responses. However, this process is computationally expensive and can get stuck on unreliable answers. To address this, we propose ECHO, a new method that intelligently controls how the model explores different reasoning paths. ECHO avoids wasteful or unstable branches during generation and adapts its learning strategy to focus on more trustworthy answers. Experiments show that ECHO consistently improves reasoning accuracy across multiple benchmarks, especially when computational resources are limited.