LGAIApr 21

LEPO: Latent Reasoning Policy Optimization for Large Language Models

arXiv:2604.1789296.4h-index: 6
AI Analysis

This work addresses the limitation of deterministic latent reasoning in LLMs by enabling exploration, which is crucial for RL-based fine-tuning.

LEPO introduces controllable stochasticity into latent reasoning for LLMs via Gumbel-Softmax, enabling diverse trajectory sampling and direct RL optimization on continuous latent representations, achieving significant improvements over existing methods.

Recently, latent reasoning has been introduced into large language models (LLMs) to leverage rich information within a continuous space. However, without stochastic sampling, these methods inevitably collapse to deterministic inference, failing to discover diverse reasoning paths. To bridge the gap, we inject controllable stochasticity into latent reasoning via Gumbel-Softmax, restoring LLMs' exploratory capacity and enhancing their compatibility with Reinforcement Learning (RL). Building on this, we propose \textbf{\underline{L}}atent R\textbf{\underline{e}}asoning \textbf{\underline{P}}olicy \textbf{\underline{O}}ptimization~(\textbf{LEPO}), a novel framework that applies RL directly to continuous latent representations. Specifically, in rollout stage, LEPO maintains stochasticity to enable diverse trajectory sampling, while in optimization stage, LEPO constructs a unified gradient estimation for both latent representations and discrete tokens. Extensive experiments show that LEPO significantly outperforms existing RL methods for discrete and latent reasoning.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes