返回全部动态
SELR：自解释潜在推理框架，兼顾效率与可解释性
原标题：Think in Latent, Explain in Language: Self-Explainable Latent Reasoning
AI 摘要
arXiv 论文提出 SELR 框架，通过多任务训练让单一模型同时进行高效潜在空间推理并生成可读的思维链解释，解决了潜在推理的不可解释性和监督缺失问题。实验表明 SELR 在 LLM 和 VLM 上相比基线提升了 token 效率和准确性，且无需外部解码器。
以上摘要由 AI 生成，可能存在误差。事实请以原文为准。
正文节选
Think in Latent, Explain in Language: Self-Explainable Latent Reasoning 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. 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
发布时间：2026-08-18 12:00
抓取时间：2026-08-17 12:08
来源机构：arXiv