Exploring Response Uncertainty in MLLMs: An Empirical Evaluation under Misleading Scenarios

Ensuring that Multimodal Large Language Models (MLLMs) maintain consistency in their responses is essential for developing trustworthy multimodal intelligence. However, existing benchmarks include many samples where all MLLMs \textit{exhibit high response uncertainty when encountering misleading information}, requiring even 5-15 response attempts per sample to effectively assess uncertainty. Therefore, we propose a two-stage pipeline: first, we collect MLLMs' responses without misleading information, and then gather misleading ones via specific misleading instructions. By calculating the misleading rate, and capturing both correct-to-incorrect and incorrect-to-correct shifts between the two sets of responses, we can effectively metric the model's response uncertainty. Eventually, we establish a \textbf{\underline{M}}ultimodal \textbf{\underline{U}}ncertainty \textbf{\underline{B}}enchmark (\textbf{MUB}) that employs both explicit and implicit misleading instructions to comprehensively assess the vulnerability of MLLMs across diverse domains. Our experiments reveal that all open-source and close-source MLLMs are highly susceptible to misleading instructions, with an average misleading rate exceeding 86\%. To enhance the robustness of MLLMs, we further fine-tune all open-source MLLMs by incorporating explicit and implicit misleading data, which demonstrates a significant reduction in misleading rates. Our code is available at: \href{https://github.com/Yunkai696/MUB}{https://github.com/Yunkai696/MUB}

Mitigating Modality Prior-Induced Hallucinations in Multimodal Large Language Models via Deciphering Attention Causality

Multimodal Large Language Models (MLLMs) have emerged as a central focus in both industry and academia, but often suffer from biases introduced by visual and language priors, which can lead to multimodal hallucination. These biases arise from the visual encoder and the Large Language Model (LLM) backbone, affecting the attention mechanism responsible for aligning multimodal inputs. Existing decoding-based mitigation methods focus on statistical correlations and overlook the causal relationships between attention mechanisms and model output, limiting their effectiveness in addressing these biases. To tackle this issue, we propose a causal inference framework termed CausalMM that applies structural causal modeling to MLLMs, treating modality priors as a confounder between attention mechanisms and output. Specifically, by employing backdoor adjustment and counterfactual reasoning at both the visual and language attention levels, our method mitigates the negative effects of modality priors and enhances the alignment of MLLM's inputs and outputs, with a maximum score improvement of 65.3% on 6 VLind-Bench indicators and 164 points on MME Benchmark compared to conventional methods. Extensive experiments validate the effectiveness of our approach while being a plug-and-play solution. Our code is available at: https://github.com/The-Martyr/CausalMM

Look Twice Before You Answer: Memory-Space Visual Retracing for Hallucination Mitigation in Multimodal Large Language Models

Despite their impressive capabilities, Multimodal Large Language Models (MLLMs) are susceptible to hallucinations, especially assertively fabricating content not present in the visual inputs. To address the aforementioned challenge, we follow a common cognitive process - when one's initial memory of critical on-sight details fades, it is intuitive to look at them a second time to seek a factual and accurate answer. Therefore, we introduce Memory-space Visual Retracing (MemVR), a novel hallucination mitigation paradigm that without the need for external knowledge retrieval or additional fine-tuning. In particular, we treat visual prompts as supplementary evidence to be reinjected into MLLMs via Feed Forward Network (FFN) as key-value memory, when the model is uncertain or even amnesic about question-relevant visual memories. Comprehensive experimental evaluations demonstrate that MemVR significantly mitigates hallucination issues across various MLLMs and excels in general benchmarks without incurring added time overhead, thus emphasizing its potential for widespread applicability.

Multimodal Laryngoscopic Video Analysis for Assisted Diagnosis of Vocal Cord Paralysis

This paper presents the Multimodal Analyzing System for Laryngoscope (MASL), a system that combines audio and video data to automatically extract key segments and metrics from laryngeal videostroboscopic videos for clinical assessment. MASL integrates glottis detection with keyword spotting to analyze patient vocalizations and refine video highlights for better inspection of vocal cord movements. The system includes a strobing video extraction module that identifies frames by analyzing hue, saturation, and value fluctuations. MASL also provides effective metrics for vocal cord paralysis detection, employing a two-stage glottis segmentation process using U-Net followed by diffusion-based refinement to reduce false positives. Instead of glottal area waveforms, MASL estimates anterior glottic angle waveforms (AGAW) from glottis masks, evaluating both left and right vocal cords to detect unilateral vocal cord paralysis (UVFP). By comparing AGAW variances, MASL distinguishes between left and right paralysis. Ablation studies and experiments on public and real-world datasets validate MASL's segmentation module and demonstrate its ability to provide reliable metrics for UVFP diagnosis.

Reefknot: A Comprehensive Benchmark for Relation Hallucination Evaluation, Analysis and Mitigation in Multimodal Large Language Models

Hallucination issues persistently plagued current multimodal large language models (MLLMs). While existing research primarily focuses on object-level or attribute-level hallucinations, sidelining the more sophisticated relation hallucinations that necessitate advanced reasoning abilities from MLLMs. Besides, recent benchmarks regarding relation hallucinations lack in-depth evaluation and effective mitigation. Moreover, their datasets are typically derived from a systematic annotation process, which could introduce inherent biases due to the predefined process. To handle the aforementioned challenges, we introduce Reefknot, a comprehensive benchmark specifically targeting relation hallucinations, consisting of over 20,000 samples derived from real-world scenarios. Specifically, we first provide a systematic definition of relation hallucinations, integrating perspectives from perceptive and cognitive domains. Furthermore, we construct the relation-based corpus utilizing the representative scene graph dataset Visual Genome (VG), from which semantic triplets follow real-world distributions. Our comparative evaluation across three distinct tasks revealed a substantial shortcoming in the capabilities of current MLLMs to mitigate relation hallucinations. Finally, we advance a novel confidence-based mitigation strategy tailored to tackle the relation hallucinations problem. Across three datasets, including Reefknot, we observed an average reduction of 9.75% in the hallucination rate. We believe our paper sheds valuable insights into achieving trustworthy multimodal intelligence. Our dataset and code will be released upon paper acceptance.

MarsQE: Semantic-Informed Quality Enhancement for Compressed Martian Image

Lossy image compression is essential for Mars exploration missions, due to the limited bandwidth between Earth and Mars. However, the compression may introduce visual artifacts that complicate the geological analysis of the Martian surface. Existing quality enhancement approaches, primarily designed for Earth images, fall short for Martian images due to a lack of consideration for the unique Martian semantics. In response to this challenge, we conduct an in-depth analysis of Martian images, yielding two key insights based on semantics: the presence of texture similarities and the compact nature of texture representations in Martian images. Inspired by these findings, we introduce MarsQE, an innovative, semantic-informed, two-phase quality enhancement approach specifically designed for Martian images. The first phase involves the semantic-based matching of texture-similar reference images, and the second phase enhances image quality by transferring texture patterns from these reference images to the compressed image. We also develop a post-enhancement network to further reduce compression artifacts and achieve superior compression quality. Our extensive experiments demonstrate that MarsQE significantly outperforms existing approaches for Earth images, establishing a new benchmark for the quality enhancement on Martian images.

Coverage-Guaranteed Prediction Sets for Out-of-Distribution Data

Out-of-distribution (OOD) generalization has attracted increasing research attention in recent years, due to its promising experimental results in real-world applications. In this paper,we study the confidence set prediction problem in the OOD generalization setting. Split conformal prediction (SCP) is an efficient framework for handling the confidence set prediction problem. However, the validity of SCP requires the examples to be exchangeable, which is violated in the OOD setting. Empirically, we show that trivially applying SCP results in a failure to maintain the marginal coverage when the unseen target domain is different from the source domain. To address this issue, we develop a method for forming confident prediction sets in the OOD setting and theoretically prove the validity of our method. Finally, we conduct experiments on simulated data to empirically verify the correctness of our theory and the validity of our proposed method.

Generalization Bounds for Adversarial Contrastive Learning

Deep networks are well-known to be fragile to adversarial attacks, and adversarial training is one of the most popular methods used to train a robust model. To take advantage of unlabeled data, recent works have applied adversarial training to contrastive learning (Adversarial Contrastive Learning; ACL for short) and obtain promising robust performance. However, the theory of ACL is not well understood. To fill this gap, we leverage the Rademacher complexity to analyze the generalization performance of ACL, with a particular focus on linear models and multi-layer neural networks under $\ell_p$ attack ($p \ge 1$). Our theory shows that the average adversarial risk of the downstream tasks can be upper bounded by the adversarial unsupervised risk of the upstream task. The experimental results validate our theory.