Variational Proximal Policy Optimization
About
Reinforcement Learning from Human Feedback via Proximal Policy Optimization often suffers from policy mode collapse, brittle exploration loops, and distribution drift. This paper introduces Variational Proximal Policy Optimization (\(\textsc{VP}_2\textsc{O}\)), a particle-based variational inference framework that maps policy optimization to Stein Variational Gradient Descent within a Mixture-of-Experts architecture. By leveraging functional kernels over localized expert prototypes alongside an expert orthogonalization loss, \(\textsc{VP}_2\textsc{O}\) introduces a geometry-based proximal-control mechanism that can reduce reliance on fixed clipping or KL schedules. Our results on a 33B/4B sparse Mixture-of-Experts model show several improvements across complex reasoning benchmarks, establishing a \(+\mathbf{179}\) ELO gain on Codeforces and a \(\mathbf{32\%}\) reduction in token count on AIME mathematical reasoning tasks.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Knowledge | MMLU-Pro | Score72.5 | 98 | |
| Mathematics | AIME 2024 | Accuracy77.5 | 87 | |
| Math | AIME 2025 | Accuracy (%)69.9 | 26 | |
| Science | GPQA Diamond | Accuracy67 | 19 | |
| Code Generation | CodeForces | ELO1.67e+3 | 4 | |
| Instruction Following | IFBench | Loose Prompt Success Rate72.2 | 4 |