Think in Latent, Explain in Language: Self-Explainable Latent Reasoning
Think in Latent, Explain in Language: Self-Explainable Latent Reasoning
潜空间思考，语言化解释：自解释潜空间推理
Abstract: Latent reasoning has emerged as a powerful alternative to text-based Chain-of-Thought (CoT), offering significant gains in computational efficiency by compressing verbose reasoning into compact embeddings. 摘要： 潜空间推理已成为基于文本的思维链（CoT）的一种强大替代方案，它通过将冗长的推理过程压缩为紧凑的嵌入（embeddings），显著提升了计算效率。
However, compressing reasoning into the latent space renders the thinking opaque, hindering its interpretability. Current methods present a stark trade-off: they either function as unexplainable “black boxes” (e.g., Coconut), where the latent reasoning is not human-readable, or rely on separate post-hoc decoders for explainability (e.g., Heima), introducing architectural overhead and decoupling the explanation from the actual reasoning process. 然而，将推理压缩到潜空间会使思维过程变得不透明，从而阻碍了其可解释性。目前的方法面临着严峻的权衡：它们要么表现为不可解释的“黑盒”（例如 Coconut），导致潜空间推理无法被人类阅读；要么依赖独立的后置解码器来实现可解释性（例如 Heima），这不仅引入了架构开销，还导致了解释与实际推理过程的脱节。
In this work, we present a unified framework for Self-Explainable Latent Reasoning (SELR) that trains a single model to perform efficient and inherently explainable latent reasoning. Our core contribution is a novel multi-task training objective that optimizes for two goals simultaneously: (1) an Answer Loss that optimizes the latent reasoning trajectory to produce accurate final answers, and (2) a CoT Loss that explicitly trains the same model to decode its own latent representations back into human-understandable reasoning steps. 在这项工作中，我们提出了一个自解释潜空间推理（SELR）的统一框架，通过训练单一模型来执行高效且具有内在可解释性的潜空间推理。我们的核心贡献是一种新颖的多任务训练目标，它同时优化两个目标：（1）答案损失（Answer Loss），优化潜空间推理轨迹以产生准确的最终答案；（2）思维链损失（CoT Loss），显式训练同一模型将其自身的潜空间表示解码回人类可理解的推理步骤。
This design ensures that generated latent representations are both task-effective and semantically interpretable, eliminating the need for external decoders. We validate the effectiveness of SELR on both Large Language Models (LLMs) and Vision-Language Models (VLMs), demonstrating that SELR achieves superior token efficiency and accuracy compared to baselines, while uniquely providing self-contained explainability without auxiliary models. 这种设计确保了生成的潜空间表示既具有任务有效性，又具有语义可解释性，从而消除了对外部解码器的需求。我们在大语言模型（LLMs）和视觉语言模型（VLMs）上验证了 SELR 的有效性，结果表明，与基线模型相比，SELR 在实现卓越的 Token 效率和准确性的同时，还独特地提供了无需辅助模型的自包含可解释性。