Skip to content
Conference

Entropy-Adaptive Routing Selection for Multimodal Chain-of-Thought Reasoning

Jun 2026 · 2026 11th International Conference on Electronic Technology and Information Science (ICETIS) · pp. 265-271 · 0 citations · 32 references

Abstract

Multimodal Chain-of-Thought (MCoT) endeavors to reason across diverse modalities sequentially, facing primary challenges of logical inconsistency with visual evidence and exorbitant computational overhead in long-chain inference. The prompt-based paradigm has been predominantly adopted, relying on zero-shot instructions and task orchestration to generate intermediate rationales. However, semantic inconsistencies between visual and textual contents are commonly encountered, which readily lead to flawed reasoning chains and unreliable inference results. In contrast, learning-based methods align cross-modal information to mitigate reasoning discrepancies, but suffer from substantial training overhead and insufficient labeled datasets. In this study, we introduce EARS-MCoT, a training-free framework centered on Entropy-Adaptive Routing Selection, aiming to bridge the semantic gap between visual and textual modalities while mitigating the computational burden of exhaustive multi-step reasoning. Initially, large language models generate the predictive probability of each token for tokenlevel entropy calculation. Token-level entropy is aggregated via a sliding window mechanism to generate window-level entropy for capturing holistic reasoning fluctuations. Finally, by integrating the Tree-of-Thoughts (ToT) paradigm, EARS-MCoT dynamically optimizes reasoning paths characterized by high entropy and uncertainty. Experimental results demonstrate that EARS-MCoT surpasses strong tree-free baselines on two mainstream visual question answering datasets, achieving superior robustness and efficiency in complex multimodal reasoning tasks.

View source

Similar papers

Preprint Aug 2026

ChainPrune: Evaluating and Reducing Redundancy in Long Chain-of-Thought Reasoning

This work proposes ChainPrune, a novel reasoning path semantic structural optimization method to efficiently and controllably synthesize self-generated high-quality training data and incorporates a DPO-based preference learning method combined with supervised loss, effectively mitigating false reward suppression.

Weihang Pan, Zhengxu Yu, Yuxiang Zhang et al. · 1 citation

Streamlining Long-Chain Reasoning via Differentiable Hierarchical Fusion

Differentiable Hierarchical Fusion is presented, a novel framework that merges reasoning models with efficient base models via differentiable optimization to produce concise, accurate outputs and introduces a dual-factor adaptive weighting mechanism to capture intra-block variance and inter-block importance hierarchies, thereby addressing key limitations of static merging heuristics.

Chuangen Gao, Wenlun Zhang, Shangkun Wang et al. · 0 citations
Book Open access Aug 2026

Learning Graph Rationales to Compress Long Chains of Thought in Multimodal Reasoning

Recent advances in multimodal large reasoning models (MLRMs) have demonstrated impressive capabilities on complex multimodal tasks, yet their reliance on long Chain-of-Thoughts (CoTs) often leads to redundant reasoning and high computational cost. Existing chain-based distillation and refinement approaches alleviate redundancy to some extent but remain limited when handling ultra-long reasoning chains, suffering from weakened dependency modeling, accumulated redundancy, and the omission of critical reasoning steps. To this end, in this paper, we propose GRACE, a Graph-based Rationalized CoT Compression framework that transforms long CoTs into reasoning graphs to enable efficient CoT compression, where reasoning steps are represented as nodes and semantic dependencies as edges. Specifically, after constructing the reasoning graph, GRACE applies a graph rationale learning approach with a selector–predictor architecture. Among them, the selector estimates node-level rationale scores to sample a compact rationale subgraph, while the predictor leverages this subgraph for final prediction. The resulting subgraph serves as the distilled core rationale, retaining essential reasoning while removing irrelevant steps. Finally, the compressed CoTs are then used to build supervised fine-tuning datasets for further MLRMs tuning. Experiments on multiple multimodal benchmarks show that GRACE substantially improves inference efficiency and interpretability without sacrificing accuracy.

Yizhi Wang, Linan Yue, Deng-Bao Wang et al. · 0 citations
Preprint Jul 2026

CAER: Conflict-Aware Evidence Routing with Dual Prefix Experts for Multimodal Large Language Models

CAER introduces a span-grounded evidence router that transforms claim representations into soft textual queries and retrieves corresponding evidence from frozen visual tokens, enabling fine-grained conflict estimation and design a dual-prefix expert routing mechanism that learns separate experts for visually supported and contradicted inputs, enabling conflict-aware generation through explicit expert selection.

Zixuan Liu, Juntao Cai, Xiaoxu Cai et al. · 0 citations
Conference Jul 2026

AdaptiReason: Adaptive Chain-of-Thought Compression for Efficient Large Language Model Inference

Large Language Models (LLMs) trained using Chain-of-Thought (CoT) supervision have achieved state-of-the-art performance on complex reasoning tasks. However, the generation of long reasoning chains introduces substantial computational overhead during inference, limiting their deployment in low-latency and resource-constrained environments. This paper proposes AdaptiReason, a novel framework that dynamically compresses intermediate reasoning steps based on task difficulty and model confidence without requiring retraining of the underlying base model. AdaptiReason employs a lightweight difficulty estimator to determine the appropriate reasoning depth for each input, followed by a learned token-pruning policy that eliminates redundant or low-information reasoning steps. Experimental evaluation on the MATH, GSM8K, and ARC-Challenge benchmarks demonstrates that AdaptiReason reduces the average number of generated tokens by 3.7× while preserving 98.2% of the baseline reasoning accuracy. Furthermore, the proposed framework is model-agnostic and can be seamlessly integrated with instruction-tuned LLMs without requiring access to model parameters, relying solely on output logits for adaptive reasoning compression. The results demonstrate that AdaptiReason significantly improves inference efficiency while maintaining high reasoning performance, making it suitable for real-time and resource-constrained LLM applications.

V. A, Mithaguru, Amrita Kundu et al. · 0 citations
Preprint Aug 2026

SCOUT: Unlocking Enhanced Spatial Reasoning via Structured Chain-of-Thought and Multi-Objective Process Reward

This work designs a structured Chain-of-Thought (CoT) framework that explicitly models 3D environmental perception to ensure robust spatial understanding and reasoning and introduces a novel RL algorithm featuring multi-objective process rewards and a tailored advantage estimation method, facilitating fine-grained credit assignment across distinct segments of the reasoning trajectory.

Zile Zhou, Huining Yuan, Weichen Zhang et al. · 0 citations