Scaling Reasoning Efficiently via Relaxed On-Policy Distillation

arXiv:2603.11137 · cs.LG, cs.CL · Submitted 2026-03-11 · Read on arXiv

cs.LG, cs.CL

Submitted: 2026-03-11

Updated: 2026-10-27

Terminology

• Distillation — 5× in this paper · explained in 2 episodes
• On-Policy Distillation — 5× in this paper · explained in 3 episodes
• Policy distillation — 5× in this paper · explained in 2 episodes
• Sample Efficiency — 2× in this paper · explained in 7 episodes
• Test-Time Scaling — 1× in this paper · explained in 4 episodes

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks