Are Multimodal LLMs Robust Against Adversarial Perturbations? RoMMath: A Systematic Evaluation on Multimodal Math Reasoning (2025.naacl-long)
Copied to clipboard
| Challenge: | Recent-released MLLMs have shown remarkable performance on various multimodal math reasoning benchmarks. |
| Approach: | They introduce RoMMath, the first benchmark designed to evaluate the capabilities and robustness of multimodal large language models in handling multimodal math reasoning. |
| Outcome: | The proposed model performs well on a broad spectrum of 17 MLLMs and demonstrates that they are robust to adversarial perturbations. |
Similar Papers
ErrorRadar: Benchmarking Complex Mathematical Reasoning of Multimodal Large Language Models Via Error Detection (2026.findings-acl)
Copied to clipboard
Yibo Yan, Shen Wang, Jiahao Huo, Hang Li, Boyan Li, Jiamin Su, Xiong Gao, YiFan Zhang, Tianlong Xu, Zhendong Chu, Aoxiao Zhong, Kun Wang, Hui Xiong, Philip S. Yu, Xuming Hu, Qingsong Wen
| Challenge: | Current mathematical benchmarks focus on evaluating MLLMs’ problem-solving ability, yet there is a crucial gap in addressing more complex scenarios such as error detection. |
| Approach: | They propose to evaluate multimodal error detection by evaluating two sub-tasks error step identification and error categorization. |
| Outcome: | The proposed task evaluates MLLMs' ability to handle multimodal questions compared to text-only models. |
From LLMs to MLLMs: Exploring the Landscape of Multimodal Jailbreaking (2024.emnlp-main)
Copied to clipboard
| Challenge: | Recent advances in Large Language Models (LLMs) have demonstrated remarkable performance across various tasks, effectively following instructions to meet diverse user needs. |
| Approach: | They propose a framework for evaluation benchmarks and attack techniques for LLMs and MLLMs to enhance their security. |
| Outcome: | The proposed frameworks have been exploited to exploit the weaknesses of LLMs and MLLMs. |
A Survey of Mathematical Reasoning in the Era of Multimodal Large Language Model: Benchmark, Method & Challenges (2025.findings-acl)
Copied to clipboard
Yibo Yan, Jiamin Su, Jianxiang He, Fangteng Fu, Xu Zheng, Yuanhuiyi Lyu, Kun Wang, Shen Wang, Qingsong Wen, Xuming Hu
| Challenge: | This survey provides **the first comprehensive analysis of mathematical reasoning in the era of multimodal large language models** . integrating large language model with mathematical reasoning tasks is becoming significant as AI advances . |
| Approach: | They review over 200 studies published since 2021 and examine the state-of-the-art developments in Math-LLMs . they identify five major challenges hindering the realization of AGI in this domain . |
| Outcome: | The authors examine the state-of-the-art developments in Math-LLMs with a focus on multimodal settings. |
MLLM-Bench: Evaluating Multimodal LLMs with Per-sample Criteria (2025.naacl-long)
Copied to clipboard
Wentao Ge, Shunian Chen, Hardy Chen, Nuo Chen, Junying Chen, Zhihong Chen, Wenya Xie, Shuo Yan, ChenghaoZhu ChenghaoZhu, Ziyue Lin, Dingjie Song, Xidong Wang, Anningzhe Gao, Zhang Zhiyi, Jianquan Li, Xiang Wan, Benyou Wang
| Challenge: | Existing evaluation methodologies for multimodal large language models are limited in evaluating objective queries without considering real-world user experiences. |
| Approach: | They propose to evaluate multimodal large language models with per-sample criteria using potent MLLM as the judge. |
| Outcome: | The proposed evaluation paradigm shows that it can be used to evaluate multimodal large language models with per-sample criteria. |
Math-LLaVA: Bootstrapping Mathematical Reasoning for Multimodal Large Language Models (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing image instruction fine-tuning datasets do not fully exploit visual information to enhance multimodal reasoning capabilities of Large language models (LLMs). |
| Approach: | They propose a LLaVA-based model fine-tuned with MathV360K to bridge this gap by collecting 40K high-quality images with question-answer pairs from 24 existing datasets and synthesizing 320K new pairs. |
| Outcome: | The proposed model improves the multimodal reasoning capabilities of LLaVA-1.5 and demonstrates enhanced generalizability on the MMMU benchmark. |
MM-JudgeBias: A Benchmark for Evaluating Compositional Biases in MLLM-as-a-Judge (2026.acl-long)
Copied to clipboard
| Challenge: | Multimodal Large Language Models (MLLMs) are increasingly used as automatic judges . however, their reliability and vulnerabilities to biases remain underexplored . |
| Approach: | They propose a benchmark to evaluate MLLMs that fail to integrate visual cues . they also introduce a test to evaluate the reliability of MLMLs based on a set of asymmetric evaluation tendencies. |
| Outcome: | Experiments on 26 state-of-the-art MLLMs reveal modality neglect and asymmetric evaluation tendencies . a standardized model with a benchmark enables a fine-grained diagnosis of nine bias types . |
Multimodal Inconsistency Reasoning (MMIR): A New Benchmark for Multimodal Reasoning Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing Multimodal Large Language Models (MLLMs) are predominantly trained on consistent visual-textual inputs, leaving open the question of whether they can handle semantic mismatches in layout-rich content. |
| Approach: | They propose to use multimodal inconsistency reasoning to assess MLLMs' ability to reason about semantic mismatches in webpages, presentation slides, and posters. |
| Outcome: | The proposed model outperforms open-source models in detecting inconsistencies in webpages, presentation slides, and posters while remaining vulnerable to inconsistent errors. |
Exploring Response Uncertainty in MLLMs: An Empirical Evaluation under Misleading Scenarios (2025.emnlp-main)
Copied to clipboard
Yunkai Dang, Mengxi Gao, Yibo Yan, Xin Zou, Yanggan Gu, Jungang Li, Jingyu Wang, Peijie Jiang, Aiwei Liu, Jia Liu, Xuming Hu
| Challenge: | Existing studies have focused mainly on visual–textual misalignment, leaving largely unexplored the MLLMs’ ability to preserve an original correct answer when confronted with misleading information. |
| Approach: | They propose a two-stage evaluation pipeline to quantify the response uncertainty phenomenon by eliciting each model’s original response on unperturbed inputs and injecting explicit (false-answer hints) and implicit (contextual contradictions) misleading instructions. |
| Outcome: | The proposed model overturns a correct answer in 65% of cases after receiving a single deceptive cue. |
GSM-Plus: A Comprehensive Benchmark for Evaluating the Robustness of LLMs as Mathematical Problem Solvers (2024.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) have demonstrated impressive performance across various mathematical reasoning benchmarks. |
| Approach: | They introduce an adversarial grade school math dataset and explore whether LLMs can be more robust when questions are slightly changed. |
| Outcome: | The proposed method generates and verifies each intermediate thought based on its reasoning goal and calculation result. |
Quantifying and Mitigating Unimodal Biases in Multimodal Large Language Models: A Causal Perspective (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Recent advances in Large Language Models have facilitated the development of Multimodal LLMs. |
| Approach: | They propose a causal framework to interpret unimodal biases in visual question answering problems and a framework to integrate information from different modalities and mitigate biase. |
| Outcome: | The proposed framework analyzes visual question answering (VQA) problems to assess their impact on predictions. |