Think Consistently, Reason Efficiently: Energy-Based Calibration for Implicit Chain-of-Thought

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chen, Zhikang, Cui, Sen, Ye, Deheng, Zhang, Yu, Bian, Yatao, Zhu, Tingting
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915609603735552
author Chen, Zhikang
Cui, Sen
Ye, Deheng
Zhang, Yu
Bian, Yatao
Zhu, Tingting
author_facet Chen, Zhikang
Cui, Sen
Ye, Deheng
Zhang, Yu
Bian, Yatao
Zhu, Tingting
contents Large Language Models (LLMs) have demonstrated strong reasoning capabilities through \emph{Chain-of-Thought} (CoT) prompting, which enables step-by-step intermediate reasoning. However, explicit CoT methods rely on discrete token-level reasoning processes that are prone to error propagation and limited by vocabulary expressiveness, often resulting in rigid and inconsistent reasoning trajectories. Recent research has explored implicit or continuous reasoning in latent spaces, allowing models to perform internal reasoning before generating explicit output. Although such approaches alleviate some limitations of discrete CoT, they generally lack explicit mechanisms to enforce consistency among reasoning steps, leading to divergent reasoning paths and unstable outcomes. To address this issue, we propose EBM-CoT, an Energy-Based Chain-of-Thought Calibration framework that refines latent thought representations through an energy-based model (EBM). Our method dynamically adjusts latent reasoning trajectories toward lower-energy, high-consistency regions in the embedding space, improving both reasoning accuracy and consistency without modifying the base language model. Extensive experiments across mathematical, commonsense, and symbolic reasoning benchmarks demonstrate that the proposed framework significantly enhances the consistency and efficiency of multi-step reasoning in LLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2511_07124
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Think Consistently, Reason Efficiently: Energy-Based Calibration for Implicit Chain-of-Thought
Chen, Zhikang
Cui, Sen
Ye, Deheng
Zhang, Yu
Bian, Yatao
Zhu, Tingting
Computation and Language
Artificial Intelligence
Machine Learning
Large Language Models (LLMs) have demonstrated strong reasoning capabilities through \emph{Chain-of-Thought} (CoT) prompting, which enables step-by-step intermediate reasoning. However, explicit CoT methods rely on discrete token-level reasoning processes that are prone to error propagation and limited by vocabulary expressiveness, often resulting in rigid and inconsistent reasoning trajectories. Recent research has explored implicit or continuous reasoning in latent spaces, allowing models to perform internal reasoning before generating explicit output. Although such approaches alleviate some limitations of discrete CoT, they generally lack explicit mechanisms to enforce consistency among reasoning steps, leading to divergent reasoning paths and unstable outcomes. To address this issue, we propose EBM-CoT, an Energy-Based Chain-of-Thought Calibration framework that refines latent thought representations through an energy-based model (EBM). Our method dynamically adjusts latent reasoning trajectories toward lower-energy, high-consistency regions in the embedding space, improving both reasoning accuracy and consistency without modifying the base language model. Extensive experiments across mathematical, commonsense, and symbolic reasoning benchmarks demonstrate that the proposed framework significantly enhances the consistency and efficiency of multi-step reasoning in LLMs.
title Think Consistently, Reason Efficiently: Energy-Based Calibration for Implicit Chain-of-Thought
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2511.07124