让大模型只在不确定时用软推理，提升逻辑稳定性与探索能力
SeLaR: Selective Latent Reasoning in Large Language Models
- 仅在低置信度步骤启用软嵌入，高置信度仍用离散词元
- 在五个基准上超越标准链式思维和现有无训练方法
- 适合需要稳定且多样推理路径的复杂问题求解场景
链式思维（CoT）是大模型推理的核心方法，但受限于离散词元采样的表达力。近期潜空间推理方法尝试用软嵌入（概率加权的词元嵌入混合）替代离散词元，却普遍存在两个问题：(1) 全局激活会干扰高置信度步骤，影响推理稳定性；(2) 软嵌入迅速坍缩至最高概率词元方向，限制对其他推理路径的探索。为此，我们提出 SeLaR（Selective Latent Reasoning），一种轻量级、无需训练的框架。SeLaR 引入基于熵的门控机制，仅在低置信度步骤激活软嵌入，高置信度步骤保留离散解码。同时，提出熵感知对比正则化，使软嵌入远离主导词元方向，持续促进多路径潜空间探索。在五个推理基准上的实验表明，SeLaR 始终优于标准 CoT 和最先进的无训练方法。
原文摘要 · Abstract (English)
Chain-of-Thought (CoT) has become a cornerstone of reasoning in large language models, yet its effectiveness is constrained by the limited expressiveness of discrete token sampling. Recent latent reasoning approaches attempt to alleviate this limitation by replacing discrete tokens with soft embeddings (probability-weighted mixtures of token embeddings) or hidden states, but they commonly suffer from two issues: (1) global activation injects perturbations into high-confidence steps, impairing reasoning stability; and (2) soft embeddings quickly collapse toward the highest-probability token, limiting exploration of alternative trajectories. To address these challenges, we propose SeLaR (Selective Latent Reasoning), a lightweight and training-free framework. SeLaR introduces an entropy-gated mechanism that activates soft embeddings only at low-confidence steps, while preserving discrete decoding at high-confidence steps. Additionally, we propose an entropy-aware contrastive regularization that pushes soft embeddings away from the dominant (highest-probability) token's direction, encouraging sustained exploration of multiple latent reasoning paths. Experiments on five reasoning benchmarks demonstrate that SeLaR consistently outperforms standard CoT and state-of-the-art training-free methods.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。