Singapore

Multimodal Large Language Models (MLLMs) have achieved remarkable progress in multimodal reasoning. However, they often excessively rely on textual information during the later stages of inference, neglecting the crucial integration of visual input. Current methods typically address this by explicitly injecting visual information to guide the reasoning process. In this work, through an analysis of MLLM attention patterns, we made an intriguing observation: with appropriate guidance, MLLMs can spontaneously re-focus their attention on visual inputs during the later stages of reasoning, even without explicit visual injection. This spontaneous shift in focus suggests that MLLMs are intrinsically capable of performing visual fusion reasoning. Building on this insight, we introduce **Look-Back**, an implicit approach designed to guide MLLMs to &quot;look back&quot; at visual information in a self-directed manner during reasoning. Look-Back empowers the model to autonomously determine when, where, and how to re-focus on visual inputs, eliminating the need for explicit model-structure constraints or additional input. We demonstrate that Look-Back significantly enhances the model&#39;s reasoning and perception capabilities, as evidenced by extensive empirical evaluations on multiple multimodal benchmarks.

AAAI 2026

Look-Back: Implicit Visual Re-focusing in MLLM Reasoning

visual reasoning

language and vision

reinforcement learning

### Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Message from the General Chair: 
*It is my great pleasure and honor to welcome you to the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025), held in beautiful Vienna, Austria, from July 27 to August 1, 2025. ACL2025continues our field’s tradition of excellence in scholarship, innovation, and inclusivity, and I am deeply grateful to the many volunteers who have worked tirelessly to bring this event to life.* 
[Read more](https://drive.google.com/file/d/1GI_hvOpjswAuYdUTromfeDiPpCcqidwg/view?usp=sharing)

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to ACL 2025". 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 need to log in with the **email address you registered with**. 

ACL 2025

Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Multimodal Large Language Models (MLLMs) have achieved remarkable progress in multimodal reasoning. However, they often excessively rely on textual information during the later stages of inference, neglecting the crucial integration of visual input. Current methods typically address this by explicitly injecting visual information to guide the reasoning process. In this work, through an analysis of MLLM attention patterns, we made an intriguing observation: with appropriate guidance, MLLMs can spontaneously re-focus their attention on visual inputs during the later stages of reasoning, even without explicit visual injection. This spontaneous shift in focus suggests that MLLMs are intrinsically capable of performing visual fusion reasoning. Building on this insight, we introduce **Look-Back**, an implicit approach designed to guide MLLMs to "look back" at visual information in a self-directed manner during reasoning. Look-Back empowers the model to autonomously determine when, where, and how to re-focus on visual inputs, eliminating the need for explicit model-structure constraints or additional input. We demonstrate that Look-Back significantly enhances the model's reasoning and perception capabilities, as evidenced by extensive empirical evaluations on multiple multimodal benchmarks.

poster

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.

To access this event page, you need to log in with the **email address you registered with**. 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.

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.

Stereo matching recovers 3D scene information based on the correlation between corresponding pixels. Despite impressive progress, existing methods lack sufficient correlation priors in ill-posed regions such as occlusions, detailed and reflective regions. In this paper, we propose Geometry Aware Stereo Matching Network (GEAStereo) to enhance geometric structure perception and address this issue. We adaptively incorporate the Monocular Disparity Distribution Prior into the stereo cost volume, building Mono-Stereo Fusion Volume (MSFV), which effectively captures global geometric structures and rectifies the correlation information in ill-posed regions. Furthermore, we introduce rich detail information from gradient features and construct a Detail-Aware Volume (DAV) by aggregating the group-wise cost volume under the guidance of gradient spatial attention, thus enhancing the correlation modeling in detailed structures. Jointly, MSFV and DAV provide rich correlation priors for disparity iterative optimization. Experimental results show that our method achieves competitive results on the ETH3D and KITTI2015 benchmarks. Compared with the state-of-the-art methods, our method demonstrates stronger performance in zero-shot generalization. The code is available in Supplementary Material.

Geometry-Aware Stereo Matching via Monocular Disparity Distribution Prior and Gradient Enhancement

Visual anomaly detection is limited by the lack of sufficient anomaly data. While existing anomaly synthesis methods have made remarkable progress, achieving both realism and diversity in synthesis remains a major obstacle. To address this, we propose AnomalyPainter, a novel framework that breaks the diversity-realism trade-off dilemma through synergizing Vision Language Large Model (VLLM), Latent Diffusion Model (LDM), and our newly introduced texture library Tex-9K. Tex-9K is a professional texture library containing 75 categories and 8792 texture assets crafted for diverse anomaly synthesis. Leveraging VLLM's general knowledge, reasonable anomaly text descriptions are generated for each industrial object and matched with relevant diverse textures from Tex-9K. These textures then guide the LDM via ControlNet to paint on normal images. Furthermore, we introduce Texture-Aware Latent Init to stabilize the natural-image-trained ControlNet for industrial images. Extensive experiments show that AnomalyPainter outperforms existing methods in realism, diversity, and generalization, achieving superior downstream performance.

AnomalyPainter: Vision-Language-Diffusion Synergy for Realistic and Diverse Unseen Industrial Anomaly Synthesis

Video language models (VideoLMs) have made significant progress in multimodal understanding. However, temporal understanding, which involves identifying event order, duration, and relationships across time, still remains a core challenge. Prior works emphasize positional encodings (PEs) as a key mechanism for encoding temporal structure. Surprisingly, we find that removing or modifying PEs in video inputs yields minimal degradation in the performance of temporal understanding. In contrast, reversing the frame sequence while preserving the original PEs causes a substantial drop. To explain this behavior, we conduct substantial analysis experiments to trace how temporal information is integrated within the model. We uncover a causal information pathway: temporal cues are progressively synthesized through inter-frame attention, aggregated in the final frame, and subsequently integrated into the query tokens. This emergent mechanism shows that temporal reasoning emerges from inter visual token interactions under the constraints of causal attention, which implicitly encodes temporal structure. Based on these insights, we propose two efficiency-oriented strategies: staged cross-modal attention and a temporal exit mechanism for early token truncation. Experiments on two benchmarks validate the effectiveness of both approaches.

Causality Matters: How Temporal Information Emerges in Video Language Models

Infrared and Visible Image Fusion (IVIF) integrates complementary information from distinct modalities to enhance image quality. However, the effectiveness declines under unseen conditions such as novel weather or scenes, due to domain shifts primarily from variations of data distribution in the visible modality, while the infrared modality remains relatively stable. To overcome domain shifts caused by the imbalance between modalities during image fusion, we propose a Domain Adaptation Guided Infrared and Visible Image Fusion method, termed DAFusion, leveraging a dual-rank domain adapter to enable fast adaptation to diverse adverse conditions during image fusion. Specifically, trainable low-rank and high-rank embedding spaces are respectively used to capture knowledge common across domains (domain-shared) and those unique to target domains (domain-specific). To leverage the dual-rank adapter more effectively, we develop a homeostatic knowledge allotment strategy to integrate the distinct types of knowledge dynamically based on the uncertainty value of target domains. Since domain adaptation typically optimizes for feature alignment across domains and emphasizes invariance rather than preserving specific cues critical for image fusion, while the fusion objective requires retaining discriminative and complementary features, a conflict between the two modules appears. To reconcile this, we further adopt a bi-level optimization framework that structurally decouples the two objectives, enabling the fusion module to steer the adaptation process while benefiting in return from domain-aligned representations. Experimental results on three benchmarks demonstrate that our method significantly outperforms state-of-the-art approaches, achieving both an enhancement in fusion quality and an improvement on subsequent high-level tasks.

Domain Adaptation Guided Infrared and Visible Image Fusion

Achieving a balance between low parameter count, reduced FLOPs, and high accuracy and throughput remains a central challenge in neural network design. To address this, we propose the partial channel mechanism (PCM), which leverages the inherent redundancy in feature map channels. PCM divides feature map channels into multiple groups, each processed by distinct operations such as convolution, attention, pooling, or identity mapping. Building on this, we introduce partial attention convolution (PATConv), a novel module that efficiently fuses convolution and visual attention within a unified framework. Our results demonstrate that PATConv can fully replace both standard convolution and visual attention modules, leading to significant reductions in parameters and FLOPs. Furthermore, PATConv enables three efficient visual attention variants: Partial Channel Attention, Partial Spatial Attention, and Partial Self-Attention. To further optimize the allocation of channel splits, we propose dynamic {partial convolution (DPConv), which adaptively learns the optimal split ratio for each layer, achieving a better trade-off between speed and accuracy. By integrating PATConv and DPConv, we develop a new hybrid network family, PartialNet, which achieves superior top-1 accuracy and inference speed on ImageNet-1K, and demonstrates strong performance on COCO detection and segmentation tasks.

PartialNet: Compute Less, Perform Better

The growing prevalence of tampered images poses serious security threats, highlighting the urgent need for reliable detection methods. Multimodal large language models (MLLMs) demonstrate strong potential in analyzing tampered images and generating interpretations.
However, they still struggle with identifying micro-level artifacts, exhibit low accuracy in localizing tampered text regions, and heavily rely on expensive annotations for forgery interpretation.
To this end, we introduce TextShield-R1, the first reinforcement learning based MLLM solution for tampered text detection and reasoning.
Specifically, our approach introduces Forensic Continual pre-training, an easy-to-hard curriculum that well prepares the MLLM for tampered text detection by harnessing the large-scale cheap data from natural image forensic and OCR tasks.
During fine-tuning, we perform Group Relative Policy Optimization with novel reward functions to reduce annotation dependency and improve reasoning capabilities. 
At inference time, we enhance localization accuracy via OCR Rectification, a method that leverages the MLLM’s strong text recognition abilities to refine its predictions.
Furthermore, to support rigorous evaluation, we introduce Text Forensics Reasoning (TFR) benchmark, comprising over 45k real and tampered images across 16 languages, 10 tampering techniques, and diverse domains. Rich reasoning-style annotations are included, allowing for comprehensive assessment. Our TFR benchmark simultaneously addresses seven major limitations of existing benchmarks and enables robust evaluation under cross-style, cross-method, and cross-language conditions.
Extensive experiments demonstrate that TextShield-R1 significantly advances the state of the art in interpretable tampered text detection.
Our dataset and code will be made publicly available.

TextShield-R1: Reinforced Reasoning for Tampered Text Detection

Propelled by the breakthrough in deep generative models, audio-to-image generation has emerged as a pivotal cross-modal task that converts complex auditory signals into rich visual representations. However, previous works only focus on single-source audio inputs for image generation, ignoring the multi-source characteristic in natural auditory scenes, thus limiting the performance in generating comprehensive visual content. To bridge this gap, we propose a method called MACS to conduct multi-source audio-to-image generation. To our best knowledge, this is the first work that explicitly separates multi-source audio to capture the rich audio components before image generation. MACS is a two-stage method. In the first stage, multi-source audio inputs are separated by a weakly supervised method, where the audio and text labels are semantically aligned by casting into a common space using the large pre-trained CLAP model. We introduce a ranking loss to consider the contextual significance of the separated audio signals. In the second stage, effective image generation is achieved by mapping the separated audio signals to the generation condition using only a trainable adapter and a MLP layer. We preprocess the LLP dataset as the first full multi-source audio-to-image generation benchmark. The experiments are conducted on multi-source, mixed-source, and single-source audio-to-image generation tasks. The proposed MACS outperforms the current state-of-the-art methods in 17 out of the 21 evaluation indexes on all tasks and delivers superior visual quality.

MACS: Multi-source Audio-to-image Generation with Contextual Significance and Semantic Alignment

Facial Attribute Recognition (FAR) holds significant potential for wide-ranging applications. However, traditionally trained FAR models exhibit unfairness, largely due to data bias—where certain sensitive attributes correlate statistically with target attributes. To address this, we propose a group-attention mechanism: first, each image is categorized into subgroups (e.g., Male/Female\&short hair, Male/Female\&long hair). Within the attention mechanism, distinct Query parameters are used for each group, with shared Key and Value parameters. As group-specific Query parameters are trained on subgrouped data, the noted bias is effectively mitigated. Consequently, integrating this Group-Attention into Vision Transformer (ViT) yields our novel Group-Decoupled ViT (GD-ViT) model. Moreover, to further attenuate the statistical correlation between sensitive and target attributes, we propose a Mask-Guided Correlation Suppression learning strategy. Specifically, in Stage 1, it first leverages a min-max dual-loss optimization strategy to train GD-ViT in capturing key regions related to sensitive attributes yet irrelevant to target attributes. Then, in Stage 2, it trains another GD-ViT by masking sensitive regions identified in Stage 1, fusing the masked output (as intermediate input) with the model’s intermediate outputs. This weakens regions associated with sensitive attributes while enhancing others, suppressing the learning of key features related to sensitive attributes. Consequently, it encourages the model to focus more on intrinsic target attribute regions and balances the learning process between the sensitive attribute and the target attribute. Extensive experiments demonstrate that our method achieves superior performance across three benchmark datasets for fair facial attribute recognition.

Fair Facial Attribute Recognition via Group-Decoupled Vision Transformer with Mask-Guided Correlation Suppression

Cross-Domain Few-Shot Learning (CD-FSL) remains a significant challenge due to substantial distribution shifts between source and target domains. While prior approaches primarily focus on spatial alignment, they often overlook discrepancies in the frequency domain. In this paper, we reveal frequency band discretization as a key phenomenon, characterized by intra-domain low-frequency dominance, inter-domain amplitude divergence, and limited high-frequency variation. This spectral disharmony biases models toward low-frequency components, leading to spectral collapse. We quantify spectral collapse via the effective rank, a principled measure of spectral diversity. To mitigate spectral collapse, we propose Harmonized Amplitude Perturbation (HAP), a frequency-domain augmentation strategy that perturbs the amplitude spectrum via frequency-aware gains sampled from Harmonized Distributions, while fixing the phase spectrum to maintain semantic integrity. Extensive experiments on both Cross-Domain Few-Shot Image Classification and Object Detection benchmarks demonstrate that HAP effectively increases spectral diversity and consistently improves generalization, outperforming state-of-the-art methods without introducing extra model complexity.

HAP: Harmonized Amplitude Perturbation for Cross-Domain Few-Shot Learning

Visual Autoregressive (VAR) models adopt a next-scale prediction paradigm, offering high-quality content generation with substantially fewer decoding steps. However, existing VAR models suffer from significant attention complexity and severe memory overhead due to the accumulation of key-value (KV) caches across scales. In this paper, we tackle this challenge by introducing KV cache compression into the next-scale generation paradigm. We begin with a crucial observation: attention heads in VAR models can be divided into two functionally distinct categories: *Contextual Heads* focus on maintaining semantic consistency, while *Structural Heads* are responsible for preserving spatial coherence. This structural divergence causes existing one-size-fits-all compression methods to perform poorly on VAR models. To address this, we propose **HACK**, a training-free **H**ead-**A**ware KV cache **C**ompression framewor**K**. HACK utilizes an offline classification scheme to separate head types, enabling it to apply pattern-specific compression strategies with asymmetric cache budgets for each category. By doing so, HACK effectively constrains the average KV cache length within a fixed budget $B$, reducing the theoretical attention complexity from $\mathcal{O}(n^4)$ to $\mathcal{O}(Bn^2)$. Extensive experiments on six VAR models across text-to-image, class-conditional, and 3D generation tasks validate the effectiveness and generalizability of HACK. It achieves up to 70\% KV cache compression without degrading output quality, resulting in both memory savings and faster inference. For example, HACK delivers a $1.75\times$ memory reduction and a $1.57\times$ speedup on Infinity-8B.

Downloads

Next from AAAI 2026

Geometry-Aware Stereo Matching via Monocular Disparity Distribution Prior and Gradient Enhancement

Similar lecture

Mitigating Visual Forgetting via Take-along Visual Conditioning for Multi-modal Long CoT Reasoning

.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

Geometry-Aware Stereo Matching via Monocular Disparity Distribution Prior and Gradient Enhancement

Similar lecture

Mitigating Visual Forgetting via Take-along Visual Conditioning for Multi-modal Long CoT Reasoning

Downloads