Search papers, labs, and topics across Lattice.
This paper introduces Surrogate Latent Policy Optimization (SLPO), a novel approach that integrates outcome-reward reinforcement learning into autoregressive latent reasoners, addressing the limitations of existing methods that rely on explicit Chain-of-Thought reasoning. By employing an empirical surrogate policy density for trajectory-level credit assignment and a correctness-supervised stopping mechanism, SLPO enables latent reasoning to scale effectively while maintaining computational efficiency. The results demonstrate that SLPO significantly enhances performance across various thinking settings, allowing for longer latent computations on more complex tasks, thereby achieving improved accuracy metrics like Pass@$k$.
SLPO enables latent reasoning models to leverage outcome-reward learning, dramatically enhancing their performance and efficiency in complex reasoning tasks.
Reinforcement learning with verifiable rewards has become the predominant recipe for eliciting test-time scaling in explicit Chain-of-Thought reasoners. Yet this scaling path remains computationally costly, since every intermediate step must be decoded as a language token. Latent reasoning instead carries intermediate computation as continuous vectors and already matches or surpasses explicit CoT at far shorter horizons. Despite this promise, latent reasoners remain largely imitation-bound, while explicit CoT has already moved past imitation via outcome-reward RL. Latent trajectories lack a tractable per-step likelihood and an adaptive stopping interface under fixed thinking budgets, so outcome rewards cannot elicit latent test-time scaling. We introduce Surrogate Latent Policy Optimization (SLPO) to bring outcome-reward RL to autoregressive latent reasoners: an empirical surrogate policy density over latent transitions for trajectory-level credit assignment, and a correctness-supervised stopping head that outcome-reward optimization refines into a variable-horizon policy. Across continuous and soft thinking settings, SLPO improves Pass@$k$ under parallel sampling and allocates longer latent computation to harder instances with higher deterministic accuracy.
Tsinghua AI