analiza aiqbench
This research introduces OPTS and TTPO, novel methods that integrate tree search into policy-gradient learning to better capture rare high-return trajectories within fixed computational budgets. By bounding gradient bias from adaptive expansion, the approach demonstrates significant improvements in tail returns for MuJoCo environments and win rates against PPO on Atari benchmarks. The work advances the frontiers of sample-efficient reinforcement learning by proving that structured search can enhance coverage without requiring action-distribution corrections.
OPTS-TTPO: Enhancing Finite-Sample Policy-Gradient Learning with Tree Search
arXiv:2609.40035v1 Announce Type: new Abstract: The policy-gradient theorem gives the exact gradient under the current policy, but finite on-policy samples may miss rare high-return trajectories. We study whether tree search improves their coverage within a fixed budget while controlling gradient bias. We introduce On-Policy Parallel Tree Search (OPTS) and Tree Trajectory Policy Optimization (TTPO) using on-policy tree trajectories, which sample new suffixes from the current policy at visited states. This needs no action-distribution correction, although branching changes state visitation. Our Branch Aggregation Lemma shows that branch-weighted tree statistics recover chain expectations when branch choices and weights are fixed before outgoing transitions are sampled. OPTS selects expansion states using estimated performance differences. Under deterministic dynamics, exact values, and max-backup advantages, the induced search policy's expected return improves monotonically with the budget. We bound the gradient bias from adaptive expansion and show that max backup assigns prefix credit to actions leading to better discovered suffixes. Against a finite chain reference, TTPG's measured bias stays near its no-branching level, while NaivePG's bias grows from 0.1251 to 0.4884. At matched budgets, reward- and value-guided OPTS improve correct-answer coverage and majority-vote accuracy over independent sampling. At matched branch counts, OPTS + TTPG gains coverage with a modest bias increase relative to Fixed-branch + TTPG. Under matched interaction or rollout budgets, OPTS-TTPO improves MuJoCo tail returns over PPO by up to 28.6%, achieves a 34-22-1 win-loss-tie record against PPO on Atari-57 under the last-100-log mean-return metric, and improves micro-averaged avg@32 and pass@32 over PPO across all four Qwen3 models.
Abstract:The policy-gradient theorem gives the exact gradient under the current policy, but finite on-policy samples may miss rare high-return trajectories. We study whether tree search improves their coverage within a fixed budget while controlling gradient bias. We introduce On-Policy Parallel Tree Search (OPTS) and Tree Trajectory Policy Optimization (TTPO) using on-policy tree trajectories, which sample new suffixes from the current policy at visited states. This needs no action-distribution correction, although branching changes state visitation. Our Branch Aggregation Lemma shows that branch-weighted tree statistics recover chain expectations when branch choices and weights are fixed before outgoing transitions are sampled. OPTS selects expansion states using estimated performance differences. Under deterministic dynamics, exact values, and max-backup advantages, the induced search policy's expected return improves monotonically with the budget. We bound the gradient bias from adaptive expansion and show that max backup assigns prefix credit to actions leading to better discovered suffixes. Against a finite chain reference, TTPG's measured bias stays near its no-branching level, while NaivePG's bias grows from 0.1251 to 0.4884. At matched budgets, reward- and value-guided OPTS improve correct-answer coverage and majority-vote accuracy over independent sampling. At matched branch counts, OPTS + TTPG gains coverage with a modest bias increase relative to Fixed-branch + TTPG. Under matched interaction or rollout budgets, OPTS-TTPO improves MuJoCo tail returns over PPO by up to 28.6%, achieves a 34-22-1 win-loss-tie record against PPO on Atari-57 under the last-100-log mean-return metric, and improves micro-averaged avg@32 and pass@32 over PPO across all four Qwen3 models.
Computation and Language (cs.CL); Machine Learning (cs.LG)
arXiv:2609.40035 [cs.CL]
(or arXiv:2609.40035v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.40035 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 346 cuvinte
Subiecte conexe
Schr\"odinger--F\"ollmer Actor--Critic: Diffusion Policy Improvement with Finite-Sample Analysis
arXiv · 30 sept. 2026
Physics-enhanced reinforcement learning for real-time optimal control of dynamical systems
arXiv · 20 iul. 2026
Enabling Token-wise Importance Sampling in Off-Policy RLVR with Proximal KL Regularization
arXiv · 14 oct. 2026
High-Probability Nash Regret for Decentralized Learning in Markov $\alpha$-Potential Games: Episodic and Fully Online Asynchronous Algorithms with Applications to Markov Congestion Games
arXiv · 15 sept. 2026