Singapore

Multimodal Large Language Models (MLLMs) are becoming integral to autonomous driving (AD) systems due to their strong vision-language reasoning capabilities. 
However, MLLMs are vulnerable to adversarial attacks—particularly adversarial patch attacks—which can pose serious threats in real-world scenarios. Existing patch-based attack methods are primarily designed for object detection models. Due to the more complex architectures and strong reasoning capabilities of MLLMs, these approaches perform poorly when transferred to MLLM-based systems.
To address these limitations, we propose PhysPatch, a physically realizable and transferable adversarial patch framework tailored for MLLM-based AD systems. PhysPatch jointly optimizes patch location, shape, and content to enhance attack effectiveness and real-world applicability. It introduces a semantic-based mask initialization strategy for realistic placement, an SVD-based local alignment loss with patch-guided crop-resize to improve transferability, and a potential field-based mask refinement method. Extensive experiments across open-source, commercial, and reasoning-capable MLLMs demonstrate that PhysPatch significantly outperforms state-of-the-art (SOTA) methods in steering MLLM-based AD systems toward target-aligned perception and planning outputs. Moreover, PhysPatch consistently places adversarial patches in physically feasible regions of AD scenes, ensuring strong real-world applicability and deployability.

AAAI 2026

PhysPatch: A Physically Realizable and Transferable Adversarial Patch Attack for Multimodal Large Language Models-based Autonomous Driving Systems

cv: large vision models

app: security

cv: adversarial attacks & robustness

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-26 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.<br><br>

To access this event page, you need to log in with the **email address you registered with**. <br>Access credentials will be sent to your email from Underline -  subject line "Welcome to AAAI 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://aaai.org/conference/aaai/aaai-26/registration/) first.

**Online Registration Form**: https://aaai.getregistered.net/conference-2026 

Registration Required

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

poster

Since high-fidelity reference images are difficult to obtain in real underwater scenes, most deep models trained by synthetic paired data cannot match real-world data exactly. In this paper, we propose an unsupervised training framework for underwater image enhancement by leveraging an iterative training strategy and quantification of specific neural units. Specifically, to eliminate the heavy color cast and distortion in the underwater images, we decompose the unsupervised image enhancement as two targeted sub-tasks, namely colorization and color compensation. First, a diffusion model is introduced for colorization to correct the green and blue color casts. Then, to intensify the learning ability of balanced color information, we introduce an extra network branch and propose a quantification mechanism for color compensation. The extra branch encodes style information from normal images into the generative model, while the quantification mechanism identifies and adjusts neural units relevant to warm colors, improving the model’s ability to learn balanced color feature representations for robust generation. In the end, through iterative training, color cast and distortion are progressively reduced, leading to a gradual improvement in the quality of the generated images. Experimental results on various widely used underwater datasets demonstrate that our approach achieves competitive performance, even when compared to recent supervised methods.

Learning Underwater Image Enhancement Iteratively Without Reference Images

Large Language Models (LLMs) often produce fluent but factually incorrect responses, a phenomenon known as hallucination. Abstention, where the model chooses not to answer and instead outputs phrases such as "I don't know", is a common safeguard. However, existing abstention methods typically rely on post-generation signals, such as generation variations or feedback, which limits their ability to prevent unreliable responses in advance. In this paper, we introduce Aspect-Based Causal Abstention (ABCA), a new framework that enables early abstention by analysing the internal diversity of LLM knowledge through causal inference. This diversity reflects the multifaceted nature of parametric knowledge acquired from various sources, representing diverse aspects such as disciplines, legal contexts, or temporal frames. ABCA estimates causal effects conditioned on these aspects to assess the reliability of knowledge relevant to a given query. Based on these estimates, we enable two types of abstention: Type-1, where aspect effects are inconsistent (knowledge conflict), and Type-2, where aspect effects consistently support abstention (knowledge insufficiency). Experiments on standard benchmarks demonstrate that ABCA improves abstention reliability, achieves state-of-the-art performance, and enhances the interpretability of abstention decisions. The source code and technical appendix are provided in the Supplementary Material.

Hallucinate Less by Thinking More: Aspect-Based Causal Abstention for Large Language Models

Accurate segmentation of ultrasound images is essential for reliable medical diagnoses but is challenged by poor image quality and scarce labeled data. Prior approaches have relied on manually designed, complex network architectures to improve multi-scale feature extraction. However, such handcrafted models offer limited gains when prior knowledge is inadequate and are prone to overfitting on small datasets. In this paper, we introduce DeNAS-ViT, a Data efficient NAS-optimized Vision Transformer, the first method to leverage neural architecture search (NAS) for ultrasound image segmentation by automatically optimizing model architecture through token-level search. Specifically, we propose an efficient NAS module that performs multi-scale token search prior to the ViT’s attention mechanism, effectively capturing
both contextual and local features while minimizing computational costs. Given ultrasound’s data scarcity and NAS’s inherent data demands, we further develop a NAS-guided semi-supervised learning (SSL) framework. This approach integrates network independence and contrastive learning within a stage-wise optimization strategy, significantly enhancing model robustness under limited-data conditions. Extensive experiments on public datasets demonstrate that DeNAS-ViT achieves state-of-the-art performance, maintaining robustness with minimal labeled data. Moreover, we highlight DeNAS-ViT’s generalization potential beyond ultrasound imaging, underscoring its broader applicability.

DeNAS-ViT: Data Efficient NAS-Optimized Vision Transformer for Ultrasound Image Segmentation

Facial Expression Recognition (FER) is crucial to human-computer interaction. Existing cross-domain FER (CD-FER) methods mainly focus on single-source closed-set scenarios, transferring knowledge from a single source domain to a target domain with identical class sets. However, CD-FER faces two real-world challenges: 1) the need to leverage information from multiple sources, leading to multi-domain shift, and 2) the necessity to recognize unseen target classes, resulting in class shift. These issues give rise to a novel and challenging task, which we define as Multi-domain Open-set FER (MO-FER). In this paper, we propose PromptEmo, a novel CLIP-based framework that leverages bilateral textual prompts to address both shifts in the MO-FER task. Leveraging the generalizability of LLM, PromptEmo constructs trainable positive prompts with LLM-generated emotion descriptions for seen classes, as well as template-derived negative prompts to enhance the reasoning for unseen classes. Then, we introduce a modal-task optimization paradigm organized from two perspectives: textual semantics and visual domains, yielding Intra-modal Space-specific Optimization (ISO) and Cross-modal Emotion-aware Interaction (CEI) strategies. ISO refines the CLIP-based textual space to ensure semantic separation between bilateral prompts and improves the latent visual space by promoting inter-domain alignment. Founded on ISO, CEI facilitates effective vision-language interactions, resulting in four joint loss terms that improve emotion recognition by shaping a domain-invariant, discriminative feature space. PromptEmo surpasses the current SOTA method by 7.7% AUC on unseen classes across four FER datasets, serving as a strong baseline for the MO-FER task.

PromptEmo: Learning Emotion with Bilateral Textual Prompts in Multi-Domain Open-set Scenarios

Color temperature, as a crucial attribute influencing image color, plays a critical role in Image Aesthetics Assessment (IAA). Yet, within the existing IAA field, little light has been shed on assessing the aesthetic quality of image color temperature. To bridge this gap, we introduce a new task: Image Color Temperature Aesthetics Assessment (ICTAA). However, this task poses the following challenge: 1) Perceptual Sensitivity: humans exhibit high sensitivity to subtle shifts in color temperature, necessitating a model to enable fine-grained discrimination; 2) Spectral Continuity: The theoretical modeling of color temperature aesthetics requires continuous labels; however, the just-noticeable-difference property of human perception makes continuous labeling infeasible, necessitating a well-designed labeling strategy. 3) Label One-Sidedness: Color temperature annotations exhibit a certain degree of ambiguity and randomness; directly converting annotation results into hard labels discards the potential preference confidence of annotators. To address the aforementioned challenge, we make the following efforts. First, we propose a multi-modal contrastive learning framework, named ICTA2Net, that models color temperature differences between image pairs while strictly controlling other visual attributes. Second, leveraging weak supervision and color temperature transitivity, we discretely sample images based on an anchor image and human perceptibility to establish contrastive relations across different color temperatures. Third, we introduce an Information Entropy-weighted Accuracy (IEA) evaluation metric to better reflect the consistency between model predictions and human preference distributions. Finally, we construct a large-scale color temperature aesthetics dataset (ICTAA240K) and a comprehensive benchmark for validation. Experiments show our method outperforms existing IAA methods on this dataset, thereby setting an effective roadmap for ICTAA.

Thinking Aesthetics Assessment of Image Color Temperature: Models, Datasets and Benchmarks

Cross-Domain Decentralized Graph Learning (CD-DGL) is a promising paradigm that enables efficient, privacy-preserving collaboration among multiple parties to unlock the value of cross-domain graph data. However, it faces two fundamental challenges. First, a critical challenge arises from the severe bias in cross-domain data, leading local models to extract deviated domain knowledge. Second, the existing domain topology shift and heterogeneous model architectures make direct model aggregation infeasible.
To address these issues, we pioneer the use of Extended Persistent Homology (EPH) to reveal and quantify the problem of domain topology shift induced by the cross-domain setting. Building on this insight, we present Decentralized Graph learning with Topology-aware knowledge Fusion (DGTF), a novel framework designed to facilitate positive topological knowledge transfer in CD-DGL. Our framework achieves this by integrating two core strategies: first, a contrastive learning-based approach to extract task-agnostic topological knowledge, and second, a topology-aware, model-independent knowledge fusion method to effectively integrate this topological information. Extensive experiments conducted under various cross-domain and model-heterogeneous settings validate the superiority and effectiveness of our proposed framework.

DGTF: Cross-Domain Decentralized Graph Learning with Topology-Aware Knowledge Fusion

Incorporating explicit reasoning rules within the latent space of language models (LMs) offers a promising pathway to enhance generalisation, interpretability, and controllability. While current Transformer-based language models have shown strong performance on Natural Language Inference (NLI) tasks, they often rely on memorisation rather than explicit rule-based generalisation. This work investigates how human-interpretable reasoning rules can be explicitly encoded within
LMs with the support of Language Variational Autoencoders (VAEs), as a mechanism for generative control. We propose a complete pipeline for learning reasoning rules within Transformer-based language VAEs. This pipeline encompasses three rule-based reasoning tasks, a supporting theoretical framework, and a practical end-to-end architecture. The experiment illustrates the following findings: Disentangled reasoning: Under explicit signal supervision, reasoning rules (viewed as functional mappings) can be disentangled within the encoder’s parametric space. This separation results in distinct clustering of rules in the output feature space. Prior knowledge injection: injecting rule-based constraints into the Query enables the model to more effectively retrieve the stored Value from memory based on Key. This approach offers a simple method for integrating prior knowledge into decoder-only language models. Moreover, we found that FFN layers are better than attention layers at preserving the separation of reasoning rules in the model's parameters.

Learning to Disentangle Latent Reasoning Rules with Language VAEs: A Systematic Study

Visual grounding, the task of linking textual queries to specific regions within images, plays a pivotal role in vision-language integration. Existing methods typically rely on extensive task-specific annotations and fine-tuning, limiting their ability to generalize effectively to novel or out-of-distribution scenarios. To address these limitations, we introduce GroundingAgent, a novel agentic visual grounding framework that operates without any task-specific fine-tuning. GroundingAgent employs a structured, iterative reasoning mechanism that integrates pretrained open-vocabulary object detectors, multimodal large language models (MLLMs), and large language models (LLMs) to progressively refine candidate regions through joint semantic and spatial analyses. Remarkably, GroundingAgent achieves an average zero-shot grounding accuracy of 65.1% on widely-used benchmarks (RefCOCO, RefCOCO+, RefCOCOg), entirely without fine-tuning. Furthermore, by substituting MLLM-generated captions with the original query texts, the accuracy at the selection stage alone reaches approximately 90%, closely matching supervised performance and underscoring the critical role of LLM reasoning capabilities. GroundingAgent also offers strong interpretability, transparently illustrating each reasoning step, thus providing clear insights into its decision-making process.

Connecting the Dots: Training-Free Visual Grounding via Agentic Reasoning

Recently, multimodal large language models (MLLMs) have achieved significant advancements across various domains, and corresponding evaluation benchmarks have been continuously refined and improved. In this process, benchmarks in the scientific domain have played an important role in assessing the reasoning capabilities of MLLMs. However, existing benchmarks still face three key challenges: \textbf{1)} Insufficient evaluation of models' reasoning abilities in multilingual scenarios; \textbf{2)} Inadequate assessment of MLLMs' comprehensive modality coverage; \textbf{3)} Lack of fine-grained annotation of scientific knowledge points. To address these gaps, we propose MME-SCI, a comprehensive and challenging benchmark. We carefully collected 1,019 high-quality question-answer pairs, which involve 3 distinct evaluation modes. These pairs cover four subjects, namely mathematics, physics, chemistry, and biology, and support five languages: Chinese, English, French, Spanish, and Japanese. We conducted extensive experiments on 16 open-source models and 4 closed-source models, and the results demonstrate that MME-SCI is widely challenging for existing MLLMs. For instance, under the Image-only evaluation mode, o4-mini achieved accuracy of only 52.11\%, 24.73\%, 36.57\%, and 29.80\% in mathematics, physics, chemistry, and biology, respectively, indicating a significantly higher difficulty level compared to existing benchmarks. More importantly, using MME-SCI's multilingual and fine-grained knowledge attributes, we analyzed existing models' performance in depth and identified their weaknesses in specific domains. For example, in questions related to ``Magnetic Field'', o4-mini correctly answered only 5 out of 33 questions, thereby fine-grainedly exposing the model's vulnerabilities. These findings highlight the urgent need to enhance the scientific reasoning capabilities of MLLMs. Code and samples are available in the Supplementary Materials.

MME-SCI: A Comprehensive and Challenging Science Benchmark for Multimodal Large Language Models

Multimedia content offers additional context for recommender systems to better understand user interests. Existing studies on multimodal recommendation primarily focus on constructing item-item semantic graphs. However, most of these methods capture only shallow semantic structures based on feature similarity and struggle to model more complex or cross-entity semantic relationships (e.g., user-item). Moreover, in these methods, collaborative signals often dominate and suppress semantic knowledge, which limits its role in representation learning. To address these issues, we propose SCALE, a novel framework that combines $\underline{S}$ubspace-aware graph $\underline{C}$onstruction and contrastive $\underline{A}$lignment for multimoda$\underline{L}$ recommendation with large languag$\underline{E}$ models. Specifically, we first use large language models and encoders to extract user and item features. Following the subspace clustering assumption, we apply the Orthogonal Matching Pursuit algorithm to mine complex semantic structures within the item-item, user-user, and user-item spaces, and integrate them into a unified semantic graph. We then perform graph convolution on both the semantic and interaction graphs, and aggregate the results for recommendation. Furthermore, contrastive losses are employed to enhance semantic fusion and alignment. Extensive experiments on five real-world datasets demonstrate that SCALE significantly outperforms state-of-the-art multimodal recommendation models, highlighting its effectiveness in modeling complex relationships and integrating semantic knowledge with collaborative signals. The source code is provided in the supplementary material.

Downloads

Next from AAAI 2026

Learning Underwater Image Enhancement Iteratively Without Reference Images

Similar lecture

Spatial-Spectral Homogeneous Attacks on Physical-World Large Vision-Language Models

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from AAAI 2026

Learning Underwater Image Enhancement Iteratively Without Reference Images

Similar lecture

Spatial-Spectral Homogeneous Attacks on Physical-World Large Vision-Language Models

Downloads