Sign InOpen Brain
arXivPaperNeeds Review

TTPO: Test-Time Policy Optimization

TTPO adapts a model at test time without labels by treating agreement and disagreement differently, reducing the damage from wrong majority-vote pseudo-labels.

arXiv · Aug 27, 2026
Open Source Open MarkdownOpen JSON
Source Summary

TTPO uses majority votes as pseudo-labels but avoids trusting them uniformly. It distills agreeing rollouts with on-policy self-distillation and penalizes disagreeing rollouts with grouped reinforcement learning, while token selection down-weights converged positions and targets confident errors.

Practical Implication

Without labels, **TTPO matched label-supervised OPSD on five benchmarks**. In test-time training it raised **Qwen3-1.7B from 38.0% to 45.2%**, suggesting adaptive inference can improve a small model without an external answer key.

Agent-Ready Context
TTPO uses majority votes as pseudo-labels but avoids trusting them uniformly. It distills agreeing rollouts with on-policy self-distillation and penalizes disagreeing rollouts with grouped reinforcement learning, while token selection down-weights converged positions and targets confident errors.

Without labels, **TTPO matched label-supervised OPSD on five benchmarks**. In test-time training it raised **Qwen3-1.7B from 38.0% to 45.2%**, suggesting adaptive inference can improve a small model without an external answer key.

The evidence is limited to competition-level mathematical reasoning in the supplied abstract. Test-time parameter updates also differ from ordinary agent retries or critique loops, so applicability to coding agents and the operational cost of per-task adaptation remain open questions.
Connected Context · Feed7 Judgment

TTPO strengthens disagreement-aware test-time adaptation by assigning different learning rules to agreeing and dissenting rollouts instead of collapsing them into one majority label. Matching a supervised comparator without labels is notable, but the evidence narrows the result to mathematical reasoning. Because TTPO updates weights per task, it occupies a more operationally demanding point in the test-time toolbox than routing, sampling, or frozen-state optimization.

DemoPSD: Disagreement-Modulated Policy Self-DistillationBoth use disagreement to control self-distillation, but DemoPSD gates teacher influence by teacher–student token disagreement while TTPO separates agreeing and disagreeing rollouts into distinct optimization treatments.GradCuit: Credit-Assigned Gradient Flow Enables Robust and Interpretable Test-Time Latent ReasoningGradCuit offers a contrasting adaptation boundary: it optimizes per-query latent states with frozen weights, whereas TTPO performs test-time policy updates.Learning When to Think: Adaptive Reasoning for Test-Time Compute AllocationAdaptive reasoning can decide how much inference compute a request receives; that routing could be a prerequisite for containing TTPO’s otherwise unresolved per-task adaptation cost.$β$-OPSD: Deriving with Policy Optimization, Training with Self-Distillationβ-OPSD tunes the balance within self-distillation during training, while TTPO extends the family toward unlabeled test-time optimization driven by rollout agreement.
Context Map
modelresearch#reasoning
Uncertainty
The evidence is limited to competition-level mathematical reasoning in the supplied abstract. Test-time parameter updates also differ from ordinary agent retries or critique loops, so applicability to coding agents and the operational cost of per-task adaptation remain open questions.