再现循环变换器：思维链变换器
Reproducing Recurrent Transformers: The CoTFormer
浏览论文内容
中文总结 AI 辅助
研究思维链变换器（CoTFormer），将思维链形式化为循环潜在计算。通过困惑度和计算效率指标评估其及变体，还扩展到受控算法设置，探究能否提升归纳推理任务的分布外泛化能力。
中文摘要 AI 辅助
思维链变换器（CoTFormer）架构将思维链形式化为一种循环潜在计算形式，保留中间状态作为可关注的表示以模仿显式推理轨迹。本文通过困惑度和计算效率指标评估CoTFormer及其结构变体，还将评估扩展到受控算法设置，以确定该循环框架是否能提升归纳推理任务的分布外泛化能力。
英文摘要
The CoTFormer architecture formalizes Chain-of-Thought as a form of recurrent latent computation, preserving intermediate states as attendable representations to mimic explicit reasoning traces. In this work, we evaluate CoTFormer and its structural variants across perplexity and compute efficiency metrics. Furthermore, we extend evaluation to controlled algorithmic settings to determine whether this recurrent framework improves out-of-distribution generalisation on inductive reasoning tasks.
发表机构
- School of Electronics and Computer Science, University of Southampton（南安普顿大学电子与计算机科学学院）
机构由 AI 辅助整理，请以论文原文为准。
