Actor-Curator: Co-adaptive Curriculum Learning via Policy-Improvement Bandits for RL Post-Training

TL;DR

ACTOR-CURATOR introduces a scalable, automated curriculum learning framework for reinforcement learning post-training of large language models, optimizing problem selection via bandit algorithms to improve performance and training efficiency.

Contribution

It formulates problem selection as a non-stationary bandit problem and derives a loss function with regret guarantees, advancing curriculum learning methods for large language models.

Findings

Outperforms uniform sampling and baseline curricula on reasoning benchmarks.

Achieves up to 80% training speedup and significant performance gains.

Demonstrates improved training stability and efficiency.

Abstract

Post-training large foundation models with reinforcement learning typically relies on massive and heterogeneous datasets, making effective curriculum learning both critical and challenging. In this work, we propose ACTOR-CURATOR, a scalable and fully automated curriculum learning framework for reinforcement learning post-training of large language models (LLMs). ACTOR-CURATOR learns a neural curator that dynamically selects training problems from large problem banks by directly optimizing for expected policy performance improvement. We formulate problem selection as a non-stationary stochastic bandit problem, derive a principled loss function based on online stochastic mirror descent, and establish regret guarantees under partial feedback. Empirically, ACTOR-CURATOR consistently outperforms uniform sampling and strong curriculum baselines across a wide range of challenging reasoning benchmarks, demonstrating improved training stability and efficiency. Notably, it achieves relative gains of 28.6% on AIME2024 and 30.5% on ARC-1D over the strongest baseline and up to 80% speedup. These results suggest that ACTOR-CURATOR is a powerful and practical approach for scalable LLM post-training.