# MIRA: Reinforcing Multimodal Reasoning via Deceptive Contextual Augmentation

DevFeed: [MIRA: Reinforcing Multimodal Reasoning via Deceptive Contextual Augmentation](<https://devfeed.tech/articles/mira-reinforcing-multimodal-reasoning-via-deceptive-contextual-augmentation-115418.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.03087>)

Author: Zhihan Yin, Jianxin Liang, Yifeng Yao, Nonghai Zhang, and Bingyue Peng, Chunguang Qie, Huishuai Zhang, Dongyan Zhao

Published: 2026-11-03T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [rlvr](<https://devfeed.tech/topics/rlvr.md>), [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [data augmentation](<https://devfeed.tech/topics/data-augmentation.md>)

Tags: [ablation](<https://devfeed.tech/tags/ablation.md>), [annotations](<https://devfeed.tech/tags/annotations.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [error](<https://devfeed.tech/tags/error.md>), [hallucinations](<https://devfeed.tech/tags/hallucinations.md>), [learning](<https://devfeed.tech/tags/learning.md>), [logical](<https://devfeed.tech/tags/logical.md>), [mechanism](<https://devfeed.tech/tags/mechanism.md>), [mira](<https://devfeed.tech/tags/mira.md>), [model](<https://devfeed.tech/tags/model.md>), [multimodal](<https://devfeed.tech/tags/multimodal.md>), [multiple](<https://devfeed.tech/tags/multiple.md>), [policy](<https://devfeed.tech/tags/policy.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [recovery](<https://devfeed.tech/tags/recovery.md>), [reflection](<https://devfeed.tech/tags/reflection.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [rlvr](<https://devfeed.tech/tags/rlvr.md>), [robust](<https://devfeed.tech/tags/robust.md>)

## AI overview

MIRA aims to reduce multimodal models' susceptibility to misleading visual cues, which can compound into hallucinations during reasoning. It trains models with filtered deceptive visual contexts as "hard traps," and uses a group-wise reflection trigger to encourage detection and correction of conflicts between visual cues and logical constraints. The authors say this behavior becomes an intrinsic policy and that the method requires no human annotations. Across multiple reasoning and perception benchmarks, they report an average 7% improvement on in-domain benchmarks and consistent gains on vision-centric benchmarks. The supplied text does not specify the evaluation metrics or individual benchmark results.

## Source excerpt

arXiv:2611.03087v1 Announce Type: cross Abstract: While Reinforcement Learning with Verifiable Rewards (RLVR) has significantly advanced multimodal reasoning, existing frameworks suffer from a critical "perception-reasoning gap." Due to an over-reliance on seemingly relevant visual tokens, even minor perceptual perturbations can propagate through the reasoning process, leading to compounding hallucinations. To address this issue, we propose MIRA, a training framework that improves robustness to erroneous visual contexts while enhancing logical reasoning ability. By injecting filtered, deceptive visual contexts during training, we construct challenging "hard traps" that stress-test the model's robustness to misleading information. Furthermore, we introduce a group-wise reflection-triggering mechanism that enables the model to autonomously detect and rectify inconsistencies between visual cues and logical constraints, which is ultimately internalized as an intrinsic policy behavior. Importantly, MIRA does not rely on human annotations. Experimental results across multiple reasoning and perception benchmarks show that MIRA yields an average 7% improvement on in-domain benchmarks, with consistent improvements on vision-centric benchmarks. These findings highlight that active error recovery is essential for robust multimodal reasoning. Comprehensive ablation studies and analyses further provide insights into why MIRA is effective.