MARS: Multimodal Adaptive Reasoning Model for Avoiding Overthinking
Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (41) · 34539-34547
DOI: 10.1609/aaai.v40i41.40753 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Tan Yue und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Multimodal Large Language Models (MLLMs) have shown advanced performance in vision-language tasks. However, existing multimodal reasoning models often suffer from excessive reasoning steps, leading to high computational costs and inefficiency. In this paper, we propose the Multimodal Adaptive Reasoning Model (MARS), which enables adaptive adjustment of the reasoning strategy based on question difficulty. Specifically, MARS adopts a three-stage training framework based on our constructed training dataset (MART): 1) CoT Masking Learning to enhance reasoning logicality by predicting masked reasoning steps. 2) Adaptive Reasoning Instruction Learning to train the model to skip or keep reasoning steps according to difficulty levels. 3) CoT Lightweight Reinforcement Learning with the Information Bottleneck Principle based GRPO algorithm to reduce CoT length while maintaining performance and generalizability. Results on both in-domain and out-of-domain datasets show that MARS significantly reduces the CoT length (90.2% decrease) while improving accuracy (0.54%), outperforming existing SOTA open-source and proprietary MLLMs.
Zitieren
APA
Yue, T., Wu, Q., & Zhao, D. (2026). MARS: Multimodal Adaptive Reasoning Model for Avoiding Overthinking. Proceedings of the AAAI Conference on Artificial Intelligence, 40(41), 34539-34547. https://doi.org/10.1609/aaai.v40i41.40753
Vancouver
Yue T, Wu Q, Zhao D. MARS: Multimodal Adaptive Reasoning Model for Avoiding Overthinking. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(41):34539-34547. doi:10.1609/aaai.v40i41.40753
BibTeX
@article{Yue2026,
author = {Yue, Tan and Wu, Qiong and Zhao, Dongyan},
title = {MARS: Multimodal Adaptive Reasoning Model for Avoiding Overthinking},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
year = {2026},
volume = {40},
number = {41},
pages = {34539--34547},
doi = {10.1609/aaai.v40i41.40753}
}
EUVIMED