Cross-Modal Attention

Vision-language models (VLMs) are fragile under image corruption. We find that the wording of the question affects VLMs in two opposite ways. Verbose questions make VLMs substantially more robust---e.g., rephrasing "Is there a cat?" into "Please look carefully and answer: is there a cat?". Conversely, VLMs become more fragile under corruption when the question is semantically complex or finer-grained, e.g., "what colour is the cup left of the chair?" instead of "is there a cup?". Both effects stem from question-conditioned cross-modal attention, which induces a spectral filter over image patches: verbose questions broaden its frequency support, while fine-grained questions concentrate it onto fewer visual scales. The model's answer drifts most when this filter and the corruption sit on the same spatial frequencies. We test the filter view on Qwen3-VL and LLaVA-OneVision across GQA and CLEVR; verbose paraphrasing reduces drift variance by 70--81% on the 8B models. The practical recipe---pad the prompt---further yields measurable gains in accuracy, even under image corruption.

Multimodal large language models (MLLMs) require substantial computation to process numerous visual tokens across all transformer layers. Most methods for efficient MLLM inference exploit horizontal redundancy by compressing visual tokens. Beyond token reduction, recent studies exploit vertical redundancy through early exit or fixed-layer skipping. However, we find that the extent and distribution of this redundancy vary across inputs and differ between self-attention and MLP modules. Motivated by these observations, we propose AdaVSkip, which equips each layer with two lightweight routers that independently determine whether visual tokens pass through by or skip the self-attention and MLP modules. These decisions collectively define an input-specific visual-computation path, but their discrete and non-differentiable nature makes learning effective paths challenging. To address this challenge, we develop a progressive two-stage training framework that updates only the routers while keeping the backbone frozen. Stage I establishes an initial routing policy through supervised training with input-specific targets derived from module-wise necessity scores. To further align the routing policy with task performance, Stage II uses reinforcement learning to optimize routing decisions with direct feedback from generated answers. It combines an answer correctness reward with a skip-consistency reward that discourages excessive retention of visual-token computation. Across three MLLM backbones, AdaVSkip maintains strong task performance with substantially less computation. On LLaVA-NeXT-7B, AdaVSkip reduces FLOPs by 53.2% while preserving the original model's average performance. Combining it with visual token compression increases this reduction to 91.2%, while retaining 97.2% of the original performance on average.

Early detection of cognitive decline from speech offers a scalable and non-invasive alternative to conventional clinical assessment. Verbal fluency tasks are particularly informative, but most automated approaches aggregate features across an entire recording, overlooking temporal speech dynamics. We propose the Cognitive Instability-Aware Cross-Modal Attention Network (CCMAN), a transfer learning framework that learns task-agnostic cognitive speech representations from multiple memory-probing tasks before fine-tuning on a minute-long semantic and phonemic verbal fluency task. CCMAN integrates semantic, acoustic, and linguistic information through bidirectional cross-attention, gated multimodal fusion, and transformer-based temporal modelling to derive interpretable biomarkers of cognitive decline. Experiments were conducted on 165.44 hours of speech from 843 participants (498 healthy controls, 245 with mild cognitive impairment, and 100 with dementia). CCMAN achieved Macro-F1 scores of 0.81 and 0.59 for binary and multiclass semantic fluency classification, and 0.77 and 0.53 for phonemic fluency, consistently outperforming strong static and temporal baselines. Statistical analyses showed that semantic drift variance and pause variance, but not mean semantic drift, were significantly elevated in both MCI and dementia relative to healthy controls, while pause duration increased progressively over the task with the steepest slope in dementia, supporting global and progressive temporal speech instability as interpretable biomarkers. Evaluation on the independent PROCESS-2 benchmark further demonstrated the generalisability of the proposed framework, improving the baseline Macro-F1 by up to 9%. These findings support temporal speech instability as a dynamic speech biomarker for robust, interpretable, and generalisable early detection of cognitive decline.

Multimodal brain state decoding has largely focused on fusing paired modalities for prediction, but has rarely explored how their correspondence can be further exploited to enrich training data and improve multimodal representation learning. To address this gap, we propose CoMA-DiT, a bidirectional cross-modal Diffusion Transformer for latent augmentation that treats paired modalities as sources of mutual generative supervision rather than merely as inputs to be fused. CoMA-DiT conditions velocity prediction on the paired modality through cross-modal attention and adaptively injects the resulting variation via a reliability-gated residual mechanism. Experiments on multimodal auditory attention decoding and emotion recognition showed that CoMA-DiT consistently outperformed 20 representative baselines, achieving absolute gains of 4.28% and 6.70% in accuracy and macro-F1 over the no-augmentation baseline, respectively. Extensive ablation, sensitivity, visualization, and interpretability analyses further demonstrated its robustness, generalizability, and ability to capture functionally relevant cross-modal interactions. These findings support a broader view of multimodal learning: Paired modalities can serve not only as inputs for fusion but also as supervision sources that augment one another.

Audio-video diffusion models rely on cross-modal attention to coordinate text, sound, and visual content, yet this same mechanism can introduce subtle and systematic semantic leakage. We study these models by probing and analyzing the ``attention triangle,'' comprising the three cross-attention edges connecting the text, audio, and video streams, and examine how semantic information is routed across modalities during generation. Our analysis reveals that routing along the audio-video edge is bidirectional: audio can influence video generation, while video can influence audio generation. This edge is shaped by biases encoded in the model's parameters and emerges as a major contributor to leakage: when prompts are in tension with learned priors, cross-modal interactions may override the intended conditioning and reroute semantics toward visually canonical but incorrect outcomes. These effects suggest that semantic artifacts arise not merely from attention spreading beyond its intended target, but from structured, bias-driven interactions along specific pathways. Building on this perspective, we extract attention-derived signals that expose how semantics are distributed and grounded across modalities, and use them as a diagnostic tool to both analyze and deliberately incur leakage under controlled conditions. This enables us to probe the internal dynamics of cross-modal routing and isolate the role of individual interactions. We further leverage these signals to guide inference-time interventions that encourage more consistent cross-modal alignment. Extensive experiments support our analysis and demonstrate improved semantic grounding while preserving generation quality.

Despite recent advances in large vision-language models (LVLMs), object hallucination remains a major barrier to their reliable deployment. Existing detection methods often characterize visual grounding using attention from individual layers, leaving its evolution across layers underexplored. We propose CADMP, a lightweight object hallucination detection framework that combines adjacent-layer cross-modal attention drift with prediction sensitivity to targeted visual masking. During decoding, CADMP quantifies distributional changes between consecutive cross-modal attention maps to capture abrupt transitions in visual grounding. It then selects the transition with the largest drift, locates the corresponding visually relevant regions, and measures the change in prediction probability after masking these regions. These two signals provide complementary evidence: attention drift characterizes the stability of internal visual grounding, while probability variation verifies whether a prediction truly depends on the identified visual evidence. A lightweight detector integrates both signals to identify hallucinated predictions. Experiments on multiple benchmarks and representative open-source LVLMs demonstrate that CADMP achieves consistently competitive detection performance. Ablation studies further confirm the complementary contributions of adjacent-layer drift modeling and mask-based grounding verification.

Multimodal large language models (MLLMs) have extended the capability of large language models (LLMs) to process more contextual multimodal information, showing remarkable progress in diverse realistic multimodal applications. Despite their strong perception and reasoning abilities, recent studies reveal that MLLMs remain highly vulnerable to adversarial inputs, especially those targeting visual components. However, existing attacks mainly focus on global perturbations, lacking an understanding of how MLLMs internally interpret visual structures. In this paper, we make the attempt to investigate the intrinsic focus of MLLMs in the frequency domain and discover that their predictions are particularly sensitive to phase information, which encodes essential structural and semantic cues. Based on this observation, we propose a novel phase-aware adversarial attack framework that explicitly restricts adversarial perturbations to structure-relevant phase regions to suppress the MLLMs' focus for effective and imperceptible attacks. To further amplify the structural influence, we also introduce an auxiliary adversarial prompt learning module to guide multimodal misalignment around phase-sensitive regions, misleading the MLLM's attention toward targeted structural patterns. Extensive experiments on multiple representative MLLM models and datasets demonstrate the superior effectiveness of our method compared to existing attacks.

Recent video generators often omit audio or synthesize it in a separate stage, limiting reciprocal modeling of visual dynamics and acoustic events. We present DreamX-Creator 1.0, a compact native joint audio-video generation system centered on a 7B generator. Conditioned on a first frame and a text prompt, the generator jointly denoises modality-specialized audio and video streams. The streams are processed independently in the first half of the network and coupled in the latter half through Gated Cross-Modal Attention, whose token- and head-wise output gates modulate each active cross-modal attention-head output. A unified Audio-Video Data System constructs and filters temporally coherent clips, produces structured multimodal annotations, and organizes clips into capability-oriented data pools. Progressive Joint Training comprises two audio-video pre-training stages followed by High-Quality Finetuning. Audio-Video Reinforcement Learning further post-trains the generator with Modality-Aware Multimodal Feedback that routes video-, audio-, and cross-modal feedback to the corresponding streams. For high-resolution output, our Autoregressive 1-Step 2K Refinement pipeline adapts a bidirectional multi-step teacher into an autoregressive multi-step refiner and distills it into a student requiring one denoising evaluation per temporal chunk. Overall, DreamX-Creator 1.0 achieves native, synchronized audio-video generation with performance competitive with state-of-the-art open-source systems. By releasing our compact 7B generator and 2K Refiner, we seek to democratize native audio-video generation and provide an accessible foundation for future research in unified audio-video generative modeling.

Current Multimodal Large Language Models (MLLMs) can process diverse sensory inputs, yet their reasoning remains heavily biased toward a dominant modality, resulting in brittle cross-modal reasoning. We introduce C$^3$PO, a benchmark of 3,404 samples spanning video, audio, image, and text, evaluating two abilities: information composition (fusing dispersed evidence) and counterfactual conflict (resolving deliberate contradictions). C$^3$PO's paired IC/CC structure and four-tier design enable targeted diagnosis of when and why cross-modal reasoning fails. Built through a fully automatic pipeline using 25 logically grounded templates, C$^3$PO reveals that while humans achieve 88.64% accuracy, the best model (Gemini-3.1-Pro) reaches only 73.17%, with open-source models collapsing under conflict. Through attention probes, we find 86-95% of failures stem from modality dominance: models commit to one modality while ignoring contradictory evidence, concentrating 87-95% of attention on text. Mid-layer attention entropy predicts correctness-sustained exploration succeeds, premature collapse fails. The 56-point accuracy gap between equally complex templates reveals that performance depends on modalities' structural roles in conflict resolution, not combinations. These findings show multimodal perception does not guarantee robust reasoning; architectures must enable sustained cross-modal attention to avoid premature

Multimodal large language models (MLLMs) achieve strong performance across diverse vision-language tasks, but their efficiency is limited by the cost of processing numerous visual tokens. Visual token pruning can reduce this cost, but requires accurate token importance estimates. Recent studies have demonstrated that text-to-vision attention from middle language model layers can effectively guide visual token pruning, typically using attention from a predefined middle layer to select the visual tokens to retain. Two problems therefore remain. First, our analysis shows that the layer whose attention is most responsive to the question varies substantially across samples, making a fixed layer suboptimal. Second, obtaining attention from the appropriate middle layer requires processing numerous visual tokens through several language model layers, by which point considerable computation has already been spent. To address both problems, we propose Middle-layer Attention Prediction (MAP), which uses Question Contrastive Teacher Selection to identify a sample-specific teacher layer by contrasting attention under the original and reference questions, and distills attention from the selected layer into a lightweight predictor that estimates visual token importance from multi-modal input features. During inference, MAP combines the predicted importance scores with a diversity criterion to prune visual tokens before the first language model layer. Thus, MAP requires no attention maps for pruning and remains compatible with existing inference acceleration techniques. Across ten benchmarks on LLaVA-NeXT-7B, MAP retains 97.5% of the unpruned model performance with only 5.56% of the visual tokens, yielding a 3.09x end-to-end speedup.

Traditional reaction yield prediction is constrained by 1D quantum descriptors that lack explicit spatial information. To address this gap, a dual-modal Vision Cross-Attention architecture is proposed, fusing tabular physical-organic data with 2D molecular topologies. Notably, it is demonstrated that a generic computer vision backbone processing simple 2D skeletal structures independently outperforms purely quantum-based baselines. By synergizing both modalities, superior predictive accuracy compared to traditional methodologies is achieved by the optimal cross-attention framework (Test RMSE = 5.27%). Through mechanistic probing, active, descriptor-guided spatial querying is observed, effectively offloading macroscopic steric identification to the visual pathway. Furthermore, a dynamic chemical hierarchy is learned by the network to heavily prioritize critical steric bottlenecks, such as the aryl halide. Concurrently, residual skip connections are utilized to protect non-spatial electronic parameters from destructive attenuation during fusion. Collectively, a scalable and highly interpretable blueprint is provided for augmenting physical chemistry with deep visual learning.

Camouflaged Object Detection (COD) aims to identify and segment camouflaged objects in complex environments, which are often concealed because their color and texture are similar to the background. Several existing COD methods introduce depth maps to boost detection performance via learning complementary RGB-D features, ignoring modality-specific characteristics of concealed objects in the depth domain. To address this issue, we propose a depth collaborative network, called VCP-DCN, to mine distinguishable multi-modality features beyond visual concealed prototype in depth domain. Specifically, VCP-DCN progressively performs multi-modality alignment, interaction, and fusion for the COD task. In the \textbf{alignment} stage, we propose a Separable Prototype Embedding (SPE) module to learn modality-consistency and modality-specific RGB/depth prototype tokens through prototype contrastive learning. Furthermore, we develop a Multi-modality Dual Attention (MDA) module to enhance the cross-modal feature representation through local response maps between modality-consistency RGB/depth prototype tokens and visual tokens on the \textbf{interaction} stage. Finally, we design a Depth Adaptive Injection (DAI) module to adaptively measure contribution of RGB/depth features with a decision-making mechanism, which calculates similarity distance between RGB/depth modality-specific prototype tokens and modality-consistency ones on the \textbf{fusion} stage. Extensive experiments demonstrate the effectiveness of our VCP-DCN on three authoritative datasets.

Recent multimodal large language models (MLLMs), such as Qwen2.5-VL and InternVL3, generate large numbers of vision tokens for high-resolution inputs, leading to substantial computational cost. Existing vision token pruning methods either depend on cross-modal attention and cannot prune before the prefill stage, or rely on diversity estimation with high computational overhead. We observe that attention scores from both vision and text tokens peak at modality separator tokens, suggesting that these separators bridge the two modalities. Based on this observation, we propose SepPrune, an efficient, training-free, plug-and-play pruning method that uses the separator token as a unified query to rank and select informative vision tokens. SepPrune reuses the LLM's built-in projection parameters and requires no architectural changes. Experiments on Qwen2.5-VL-7B show that SepPrune achieves state-of-the-art performance, retaining 96.3% of the original accuracy while removing 80.2% of vision tokens.

The rapid advancement of video generation models has led to the increasing misuse of image-to-video (I2V) models. Although substantial progress has been made in detecting AI-generated videos, proactive defenses against I2V models remain underexplored. In particular, current proactive defenses against I2V models predominantly rely on gradient-based adversarial attacks, which require defenders to possess GPUs with substantial memory resources (VRAM) to generate adversarial examples. To address this issue, we propose I2VShield, a privacy protection method based on generative adversarial attacks tailored to Diffusion Transformer (DiT)-based I2V models. The proposed method primarily consists of two components: (1) a text-adaptive perturbation generation framework integrating adversarial learning to mitigate computational overhead while maintaining visual imperceptibility; and (2) an untargeted Multimodal Attention Disruption (MAD) attack that exploits the inherent vulnerabilities of DiT-based I2V models, maximizing the deviation of the internal attention features from their clean states. Extensive experiments demonstrate that our approach achieves highly competitive protection performance across various datasets and mainstream DiT-based I2V models, particularly in disrupting spatiotemporal coherence, while substantially reducing computational costs.

Real-time long-form digital-human generation relies on causal models to extend audio-visual content while preserving subject appearance and audio-video synchronization across successive segments. A bounded cache retains local motion and phonetic context but discards older evidence, whereas attending to the complete generated history is computationally expensive and can propagate accumulated errors. We present \method, an anchor-guided persistent-memory framework for few-step joint audio-video generation. The framework preserves an immutable visual anchor, compresses completed video and audio blocks into fixed-capacity dynamic states, and retrieves those states through modality-specific residual attention without extending the active cache. A reference-aware modulation method additionally conditions video features on dynamic and anchor appearance statistics. Anchor-preserving causal-context distillation varies rollout horizon, prefix provenance, and cache-history reliability while keeping the immutable visual anchor unperturbed. By separating persistent memory from stage-local denoising dependencies, \method further admits stage-parallel execution across blocks, accelerating autoregressive inference without pipeline-specific retraining. We evaluate long-form video continuations with appearance, temporal, synchronization, facial, and speech diagnostics. Results show that \method preserves stable appearance across prompt-conditioned segments and strong audio-visual synchronization under autoregressive generation. Our project page is https://taoliveaigc.github.io/TaoMate.

The empirical success of attention mechanism in Multimodal Large Language Models (MLLMs) often obscures its inherent, subtle flaws. Specifically, MLLMs consistently exhibit disproportionate attention toward certain semantically uninformative visual tokens, a phenomenon termed "register" or "Visual Attention Sinks." While existing inference intervention methods attempt to identify these sink tokens and redistribute their attention weights, such approaches typically treat these tokens in isolation and suffer from computational inefficiency. Instead, we reframe this phenomenon as a generalized textual bias exerted over visual features that extends beyond isolated sink tokens. From this perspective, a pervasive structural bias leads to the dilution of the semantic visual signal, precipitating multimodal hallucinations as the model prioritizes linguistic priors over valid visual evidence. To address this limitation, we introduce Saliency-guided Purification and Adaptive Redistribution (SPAR), a training-free, plug-and-play intervention. SPAR mitigates this generalized textual bias by purifying structural noise and subsequently redistributing the reclaimed attention budget to the most informative visual regions. Comprehensive evaluations across a diverse spectrum of hallucination benchmarks demonstrate that SPAR effectively restores authentic visual grounding with negligible computational overhead.

End-to-end autonomous driving systems commonly follow a cascaded two-stage pipeline where a perception stage compresses multi-modal sensor inputs into a compact context and a downstream planner predicts trajectories conditioned on this context. We argue that this one-way perception-to-planning interface forces sensor inputs into a compact representation, losing the fine-grained details critical for planning. Moreover, by constraining the planner to this compressed context, it is difficult to leverage the rich representations offered by modern vision foundation models. To address these issues, we propose MOJITO, a unified sensor-to-action framework for end-to-end autonomous driving built on modal joint learning. MOJITO removes the cascaded interface and instead performs block-wise Modal Joint Attention that simultaneously updates action, image, and LiDAR features, allowing the planner to directly access multi-modal features during action generation. MOJITO achieves 88.9 PDMS on the NAVSIM v1 dataset and 88.4 EPDMS on the more challenging NAVSIM v2 dataset, setting a new state-of-the-art. Extensive experiments further demonstrate strong scalability, instruction following, and diverse trajectory generation. Code and models are available at https://github.com/mumucc01/MOJITO.

Automatic depression detection with deep learning has shown promise but often suffers from limited generalization due to domain shift arising from inter-speaker variability. To address this critical issue, we present the first patient-independent multimodal depression detection framework that incorporates domain generalization (DG), jointly leveraging both acoustic and textual modalities. The proposed model integrates bidirectional Long Short-Term Memory (BiLSTM) with intra- and cross-modal attention mechanisms, accompanied by segment-level fusion for decision-making. Generalization is further enhanced by applying a gradient reversal layer inspired by Domain-Adversarial Training of Neural Networks (DANN), which promotes domain-invariant representations by adversarially limiting the model's ability to identify individual speakers, effectively reducing patient-specific bias. Conducting experiments on the Androids-Corpus dataset with a 5-fold cross-validation (CV) protocol, various pairings of audio and text feature extractors were evaluated over different segment durations, determining MelSpec and ItalianBERT as the optimal baseline at a 30-second segment duration. The addition of DG to this baseline yields a 2.5% increase in accuracy and 3.3% in F1-score, achieving 93.2% accuracy, 93.2% precision, 96.2% recall, and 94.2% F1-score, surpassing all existing benchmarks. Extensive ablation studies assess the impact of multimodal fusion, deep architectural choices, and DG, highlighting their combined contribution to robust and generalizable depression detection.

Vision-language models (VLMs) provide a unified representation space for textual and visual information, yet their potential as general-purpose backbones for graph-structured data remains largely unexplored. In practice, attributed graphs exhibit substantial modality heterogeneity: some graphs contain only textual node attributes, others only visual attributes, while still others provide both. Existing graph learning approaches are typically designed for fixed modality schemas, requiring separate models for different settings and limiting scalability and cross-graph generalization. To bridge this gap, we present OMG-VLM (One Model, Many Graphs with Vision-Language Models), a unified framework for learning over attributed graphs across heterogeneous modality schemas. OMG-VLM leverages a pretrained VLM as a shared backbone and introduces structure-aware graph adapters that integrate neighborhood information while remaining compatible with the VLM's native embedding space. This design enables effective learning over text-attributed, image-attributed, and multimodal-attributed graphs within a single model. Extensive experiments across diverse domains show that OMG-VLM consistently outperforms state-of-the-art GNN- and LLM-based baselines on attributed graph learning tasks such as node classification and link prediction, while exhibiting strong generalization to unseen graphs and varying modality schemas. The source code is available at https://github.com/Jo-eyang/OMG-VLM.

Detecting high-level semantic concepts like negation across modalities remains a challenge for current multimodal systems. We analyze this as a fundamental representation learning problem, providing the first evidence that negation does not form a linearly or non-linearly separable class in the latent spaces of standard vision-language models (VLMs). We demonstrate that pretrained embeddings primarily encode modality-specific features, lacking a generalizable negation signal. To overcome this, we propose a novel cross-modal attention architecture that explicitly models inter-modal dependencies, achieving performance gains of up to +7.03% F1 over unimodal baselines. Our analysis reveals a key asymmetry: while textual negation often appears independently, visual negation is semantically dependent on linguistic context, a finding validated through our statistical analysis of 3,222 political video-text pairs automatically annotated via \textsc{Qwen2.5-VL}. By combining this analysis with self-supervised video representations (JEPA2), we advance the modeling of temporal negation. This work provides new methods and insights for learning robust, semantically-aligned representations in multimodal systems.

Forecasting the outcomes of transition-metal-catalyzed reactions is notoriously complex due to the interplay of diverse physical and chemical variables. A persistent computational bottleneck has been effectively merging broad electronic descriptors with the localized, three-dimensional geometry of the reactive site. To bridge this representation gap, we present ChemFusion, a hybrid neural network that fuses conventional electronic features with explicit 3D atomic coordinates. Using a cross-attention mechanism, the model enables global electronic states to dynamically attend to specific spatial constraints within un-pooled molecular point clouds. When benchmarked against a diverse library of cross-couplings, this approach delivers exceptional predictive performance, decisively surpassing traditional single-modality frameworks. Importantly, extracting the attention matrices reveals that the architecture autonomously learns to identify and penalize restrictive steric hindrances. This provides a physically grounded interpretability, demonstrating that spatially aware networks can navigate complex reaction sterics that standard statistical models typically miss.

Accurate emergency triage decision is critical to avoid clinical deterioration, morbidity, and mortality. Machine learning-based triage system involves acquiring the main presenting complaint in text form and assessing vital signs in numerical data, enabling an automated and efficient analysis of patient information for timely and accurate prioritization of medical attention. However, modelling the intricacies of both data types requires a comprehensive understanding of the temporal structure and dependencies within the data. Thus, the aim of this study is to propose a multimodal deep learning architecture that can effectively handle both tabular and textual data. Furthermore, the proposed model exploits self-attention to to capture both local and global relationships between the features. A dataset consisting of 11,102 triage data collected from emergency department of Hospital Universiti Sains Malaysia is used for model development and validation. The proposed model demonstrated an increase of 1.95% in accuracy, 2.49% in F1-score, and 1.41% in ROC AUC compared to the baseline model. The experimental results demonstrated the potential of the proposed model in predicting triage decisions.

Classical approaches to event-based egomotion estimation, including those adopted by the top-performing teams of the ELOPE challenge, rely on geometric optimization frameworks such as contrast maximization, homography estimation, or dense optical flow combined with analytic motion inversion. This work investigates the geometric structure that emerges inside a multi-modal network for egomotion estimation. Event tensors, inertial measurements, and range signals are fused through a cross-modal attention architecture and trained in a batch setting. We analyze the latent space geometry and attention dynamics, showing that (i) embeddings lie on low-dimensional manifolds aligned with motion variables, (ii) attention weights adapt with angular excitation and visual reliability, and (iii) the fused representation recovers classical observability cues. These results bridge analytical estimation theory and modern data-driven fusion.

Predicting pedestrian crossing intention is a safety-critical task for autonomous driving, yet existing approaches often rely on single-modal inputs or dense multimodal fusion strategies that inadequately capture complementary visual and kinematic information while introducing redundant inter-modal interactions. We propose ADAPT (Adaptive Domain-Aware Pedestrian Crossing Transformer), a multimodal framework that jointly models local and global visual context together with temporal motion dynamics for accurate pedestrian crossing intention prediction. ADAPT processes four spatially aligned visual modalities, including RGB images, local depth maps, global semantic maps, and global depth maps, together with ego-vehicle speed, pedestrian bounding boxes, and skeleton pose information through five specialized modules: a weight-shared Swin Transformer V2 backbone for visual feature extraction, a Cross-Modality Guided Attention module for hierarchical visual fusion, a Mamba-based Motion Feature Encoding module for efficient temporal modeling, a Sparse Cross-Modal Attention module that selectively preserves the most informative inter-modal interactions, and a Vision Transformer-based Temporal Feature Fusion module for sequence-level prediction. Extensive experiments on the JAAD and PIE benchmark datasets demonstrate that ADAPT consistently outperforms existing state-of-the-art methods while maintaining low computational complexity. On JAAD, the proposed method achieves an AUC of 0.73 on JAADbeh and 0.85 on JAADall, while on PIE it achieves an accuracy of 0.92 and an AUC of 0.90. Furthermore, ADAPT performs inference in only 17.23 ms per sample, offering an effective balance between predictive accuracy and real-time deployment efficiency for intelligent transportation and autonomous driving applications.

Multimodal large language models can emit localized predictions, bounding boxes for objects and temporal windows for video and audio events, but they hallucinate these regions prolifically. The model's own token log-probabilities are nearly uninformative: they conflate grounding quality with input ambiguity, and coordinate tokens become near-deterministic once the model commits. We propose Multi-Token Localized Attention (MTLA): a training-free, post-hoc score that measures how strongly a prediction's tokens attend to the region they claim. Prior attention-based detectors, which sum attention over the entire input modality and read a single response token, are weaker special cases; we show that summing only within the claimed region and aggregating across all prediction tokens recovers a stronger grounding signal. The same recipe applies almost trivially to other modalities and tasks: object detection in images and temporal localization in video and audio. Across multiple MLLM families and three modalities, MTLA improves hallucination AUROC by +7 to +38 over the best prior training-free baseline. Used as a confidence score for re-ranking, it nearly doubles the zero-shot COCO detection AP of an open-source 8B generalist (from 20.4 to 37.0), narrowing the gap to supervised detectors without any task-specific training.

Diabetic retinopathy (DR) is a leading cause of vision impairment worldwide, highlighting the need for accurate and accessible screening tools. Optical Coherence Tomography (OCT) provides high-resolution structural information of the retina, whereas OCT angiography (OCTA) offers complementary vascular information that is highly relevant for DR diagnosis. In this study, we propose a cross-modal fusion of OCT B-scans with single-channel en face OCTA using a bidirectional cross-modal attention network for automated DR classification. Two independent datasets, OCT500 and UIC, comprising 730 subjects in total, were utilized to evaluate performance under within-dataset, combined-dataset, and cross-dataset generalization settings. A ConvNeXt V2 model trained solely on OCT images served as the unimodal baseline. In addition to ground-truth (GT) OCTA, we explored the use of translated (TR) OCTA generated from OCT scans, eliminating the requirement for dedicated OCTA hardware. Experimental results demonstrate that cross-modal fusion consistently outperforms unimodal OCT classification across all evaluation scenarios. Fusion with GT OCTA improved classification accuracy and discriminative performance, while TR OCTA achieved comparable or superior results in most settings. Furthermore, TR OCTA improved sensitivity and cross-dataset generalization, indicating enhanced robustness to domain shifts. These findings demonstrate that attention-based OCT-OCTA en face fusion provides clinically meaningful improvements for DR detection and suggest that computationally generated OCTA can serve as a practical, low-cost alternative to hardware-acquired OCTA, enabling broader deployment of high-performance retinal screening systems in resource-limited clinical environments.

Audio-Visual Question Answering (AVQA) extends classical VQA by requiring joint reasoning over video and synchronized audio. However, many AVQA systems rely on deeply stacked layers of self- and cross attention across text, video, and audio. Such sequential stacking may incur loss of information such as subtle inter-modal cues over the layers, causing errors to accumulate across sequential attention layers during the fusion. We introduce Q-TriM which performs multi-modal fusion in a shallow and parallel manner instead of a deep and sequential manner. For Q-TriM, we propose a novel framework for attention operation incorporating video and audio conditioned on text. As a result, we obtain not only standard cross attention outputs but also Tri-Modal Attention representations in which Query, Key, and Value come from distinct modalities. These attention representations are combined in parallel at a single stage, thus avoiding the multi-modal fusion with deep stacks in order to mitigate error accumulation and depth-induced issues. Q-TriM achieves state-of-the-art performance on three AVQA benchmarks, including substantial gains on MUSIC-AVQA-R, which demonstrates its robustness and out-of-distribution generalization. Code is available at https://github.com/Sunghun95/Q-TriM

Gloss-free Sign Language Translation (SLT) translates sign language videos into spoken-language sentences without gloss annotations, avoiding costly labeling but requiring fine-grained modeling of hands, body, and facial cues. Existing methods often use single-modality or weakly fused features, limiting performance. We propose ViPo-MLLM, a framework that integrates spatio-temporal RGB and human pose features. Dedicated encoders model intra-modal dynamics and cross-modal attention captures long-range dependencies. The fused representation is conditioned with a structured prompt and processed by an LLM trained with contrastive and language modeling objectives. The proposed model was evaluated on the PHOENIX14T and CSL-Daily datasets and achieved new state-of-the-art results on both datasets. Moreover, the ViPo-MLLM model attained competitive performance compared to gloss-based recognition approaches, confirming the effectiveness of the proposed pose cues and cross-modal attention mechanisms.

While knowledge distillation (KD) is widely adopted for training lightweight models by leveraging supervision from larger teacher models, relying solely on output token distributions has proven insufficient for compressing Multimodal Large Language Models (MLLMs). Since output tokens are a byproduct of the model attending to visual inputs, prior works have explored explicitly distilling attention to provide a direct supervisory signal. While promising, the precise utility of which attention signals to distill remains under-explored. In this work, we challenge the conventional reliance on prompt-to-vision attention by revealing that downstream performance correlates strongly with response-to-vision attention similarity to the teacher, but negligibly with that of prompt-conditioned attention. Furthermore, we observe that attention distributions exhibit significant variance across individual tokens, indicating that a uniform distillation objective is suboptimal. To this end, we introduce Token-level Response-visual Attention Guidance (TRAG), a distillation objective that 1) shifts the focus to response-to-vision signals and 2) employs token-specific objectives by adaptively weighting the Kullback-Leibler divergence based on attention entropy, effectively guiding the student to mirror the teacher's precise visual focus. Extensive experimental results on multiple benchmarks demonstrate that TRAG significantly outperforms prior distillation baselines.

Vision-language models (VLMs) have become a paradigm for multimodal learning, yet remain unstable due to object hallucination, weak visual grounding, and catastrophic forgetting after full-parameter instruction tuning. We claim these failures result from a lack of explicit control over visual representation learning during the standard next-token prediction objective. As a result, visual embeddings thus become passively optimized and prone to injecting redundant or spurious signals. To counter this, we introduce Information-Regularized Attention (IRA), a stochastic attention mechanism that explicitly regulates the amount of visual information injected into the hidden states of intermediate transformer layers. This local reparameterization translates uncertainty about visual representations into local noise that is independent across data points. Beyond evaluating model performance, we also quantify embedding properties, where IRA produces smoother curvature trajectories and suppresses attention-sink across all layers, indicating a more stable transformation of the visual signal. Our results suggest that stochastic attention is not merely a regularizer but a key contributor to representation learning in a generative architecture, offering a new direction for building more reliable VLMs.

Multimodal Large Language Models (MLLMs) are critically hampered by hallucination, generating content inconsistent with the provided image. In this paper, we identify an internal signature of hallucination: progressive degradation of text-to-image cross-attention during generation, leading to specific failure patterns like unfocused or biased attention. Existing mitigation strategies are largely outcome-driven and do not explicitly target this failure mode. To address this problem, we propose ADAPT (Attention Dynamics Alignment with Preference Tuning), an attention-based framework that intervenes directly on text-to-image cross-attention dynamics. We propose ADAPT with three key contributions: a cross-attention visual anchor refined from early decoding to provide stable spatial grounding, an attention-supervised inference mechanism that detects and corrects attention drift online, and a Visual Attention Guidance DPO that aligns preferences toward visually grounded responses. Experiments show that each component of ADAPT contributes to hallucination reduction, and the full framework achieves new best results across multiple hallucination benchmarks, reducing hallucination rates by 40%-60% across mainstream backbones while preserving general multimodal capabilities. Our work provides an attention-based perspective on mitigating hallucinations by exploring the model's internal text-to-image cross-attention behaviors. Code is available at https://github.com/yao-ustc/ADAPT

Sparse Autoencoders (SAEs) have shown promise for analyzing language models, but applying them to vision-language models (VLMs) often yields representations that are difficult to use as controllable cross-modal steering directions. We introduce the Joint Sparse Autoencoder (JSAE), which uses an explicit alignment constraint to jointly factorize sequence-pooled vision and language activations into shared, interpretable image/caption-level features. Applied to LLaVA, JSAE recovers cross-modal features for recognizable concepts (e.g., food and animals). Through bidirectional interventions (additive steering and suppression), we observe a layer-dependent asymmetry under our protocol: additive steering peaks at mid-to-late (pre-output) layers and weakens at both ends, whereas suppression scores remain within a comparable range across all probed layers within statistical noise. Experiments on three VLMs, namely LLaVA-v1.6-Mistral-7B, Llama3-LLaVA-8B, and the MoE-based Qwen3-VL-30B, show related layer-localized effects across architectures. Together, these results suggest that explicitly aligned sparse representations support more controllable intervention-based analysis of multimodal features, within an identifiable layer range, than the unconstrained alternatives tested here.

Background: Response to neoadjuvant imatinib in gastrointestinal stromal tumors (GISTs) is highly variable and cannot be reliably predicted using current clinical or molecular markers. This study developed and evaluated an explainable multimodal deep learning framework integrating computed tomography (CT) imaging and clinical variables to predict treatment response. Methods: Patients from four tertiary centers were retrospectively included between 2000-2023 in independent pretraining (n=935) and prediction (n=213) cohorts. A cross-attention framework integrating clinical variables and tumor-centered CT imaging was developed to predict response to neoadjuvant imatinib. Two training strategies were evaluated: (1) self-supervised pretraining with low-rank adaptation and (2) training from scratch. Hyperparameters were optimized using SMAC3. Performance was assessed through internal cross-validation and external testing. Ablation analyses and attention-based explanations were used to quantify modality contributions. Results: Among 213 patients (54.5% responders), responders had larger tumors (112 vs. 89 mm, P=0.026), higher mitotic index (3 vs. 0, P<0.001), and more frequent KIT mutations (69.0% vs. 56.7%, P=0.019). Cross-attention models achieved the highest internal performance (AUC up to 0.99) but lower external performance (AUC 0.60-0.63). Clinical-only performance was moderate (AUC 0.66), whereas imaging-only models showed limited generalizability (AUC 0.56-0.66). Explainability analyses identified significant differences in feature importance between responders and non-responders, including CD117, BRAF, PDGFRA, age, sex, disease status, and comorbidities (FDR-adjusted P<=0.036). Conclusion: The cross-attention framework shows potential for improving imatinib response prediction in GIST while providing interpretable insights into multimodal determinants of treatment response.

General-purpose embodied agents must understand multimodal instructions, anticipate how their environment will evolve, and produce precise control actions over extended horizons. Existing approaches typically specialize in visual-language reasoning, video-based world modeling, or action generation, while cascaded pipelines that first synthesize future observations and then infer actions can introduce interface bottlenecks and compound prediction errors. We present iFLYTEK-Embodied-Omni, a unified multimodal foundation model that jointly models vision(videos and images), language, and action within a single Omni framework. Its modality-specific visual-language, video-generation, and action-generation components communicate through shared multimodal self-attention. This design establishes brain-cerebellum collaboration: the vision-language modeland video generation model form a high-level brain for instruction understanding, task planning, progress tracking, and future visual-state prediction, whereas the action generation modelserves as a low-level cerebellum that directly converts planned subgoals and shared multimodal context into executable action chunks. To develop these capabilities, we combine action-annotated and action-free embodied videos from human demonstrations and robot interactions with embodied reasoning, embodied perception, and general-purpose image-text data to construct a comprehensive dataset. We further adopt a four-stage strategy that progressively trains the VLM, VGM, and AGM before jointly fine-tuning the complete model.

Reducing visual token redundancy is critical for accelerating Multimodal Large Language Models (MLLMs) without degrading cross-modal reasoning performance. Existing token pruning methods typically rely on single-layer signals, such as attention scores or token similarities, which overlook the cross-layer transformation of visual representations and may exhibit positional bias in multimodal token sequences. To address this limitation, we propose a training-free token pruning framework based on Cross-Layer Spectral Evolution (CLSE). Instead of measuring token importance from single-layer feature magnitudes, CLSE quantifies how token representations evolve across Transformer layers in the frequency domain. This evolution reflects the transition from high-frequency structural details to low-frequency semantic abstractions. We observe that tokens with stronger spectral redistribution across layers are more likely to be semantically active and should therefore be preserved. By modeling cross-layer token dynamics, CLSE provides a stable importance criterion that mitigates positional bias. Extensive experiments on both image and video benchmarks demonstrate that CLSE achieves a superior trade-off between efficiency and accuracy under aggressive token reduction. Across multiple MLLMs, CLSE reduces FLOPs, KV cache memory, and latency while maintaining competitive or improved performance.

Multimodal large language models (MLLMs) are increasingly required to adapt to non-stationary streams of visual domains, question types, and user instructions, yet continual fine-tuning often causes severe forgetting of previously acquired multimodal skills. Existing continual vision-language methods mainly preserve outputs, replay data or pseudo-data, regularize embedding geometry, or allocate task-specific parameters, but they provide limited control over how internal cross-modal attention patterns supporting old skills drift during adaptation. We propose Attention-Spectrum Regularization (ASR), a replay-free continual learning framework that preserves skill-conditioned structures of cross-modal attention. ASR treats cross-attention maps as two-dimensional signals, summarizes their scale and directional properties into compact spectral statistics, and stores only skill-wise prototype distributions instead of replaying past image-question pairs, generated pseudo-examples, or old-stage teacher snapshots. In later stages, a phase-invariant spectral regularizer constrains harmful drift of these prototypes while allowing instance-level attention to adapt to new tasks. We provide theoretical analysis showing that skill-conditioned spectral drift controls forgetting under a spectral sufficiency assumption, and that Fourier power spectra are stable to spatial translations and bounded perturbations. Experiments on continual VQA and multimodal instruction-tuning benchmarks, including VQA v2, VQACL, CLT-VQA, CoIN, and UCIT, show that ASR consistently improves final performance and reduces forgetting over strong replay-, regularization-, and adapter-based baselines. Preserving skill-level attention structure is an effective and lightweight mechanism for continual MLLMs. Code is available at https://github.com/Creative-zcx/attention-spectrum-replay

Pedestrian trajectory prediction requires modeling temporal dynamics, multimodal cues, and social interactions in crowded environments. Existing methods often address these factors separately or entangle them in costly attention blocks, limiting scalability, flexibility, and interpretability. We propose a three-step hierarchical Transformer that explicitly separates temporal encoding, multimodal fusion, and scene-level interaction reasoning. Lightweight GRU summaries enable efficient cross-modal attention, while social attention over time--agent tokens captures inter-pedestrian influences at manageable cost. Experiments on JTA, JRDB, and the Pedestrians and Cyclists in Road Traffic dataset show state-of-the-art performance on real-world datasets (JRDB, Urban) and competitive results on JTA. Ablation and qualitative analyses confirm the contribution of each stage and the model's ability to anticipate complex behaviors such as early turning.

Timely and accurate disaster damage assessment is crucial for effective emergency response, resource allocation, and recovery. Traditional methods, which often rely on manual inspections or sparse data, are typically slow and error-prone. This paper introduces a novel framework leveraging remote sensing imagery and deep learning to automate building damage classification. Using pre- and post-disaster satellite imagery, our model categorizes buildings into four damage levels: no damage, minor damage, major damage, and destroyed. The core innovation is a multi-modal attention mechanism that fuses bi-temporal features to explicitly detect and assess structural changes. We employ a lightweight ConvNeXT-Tiny backbone to ensure efficient processing without compromising performance. Key contributions include: (1) a cross-attention module for multi-modal data fusion, (2) an optimized preprocessing pipeline for large-scale datasets, and (3) robust data augmentation techniques. Experiments on a large-scale disaster dataset demonstrate an overall classification accuracy of 94.90%. The model effectively discriminates between damage categories and remains resilient to incomplete data. This system significantly improves assessment speed and accuracy, aiding emergency responders in prioritizing interventions. This work advances automated disaster damage detection by integrating multi-temporal imagery with deep learning, offering a scalable solution for real-time response.

The integration of Foundation Models (FMs) and wireless communications is driving the evolution of image communication from bit-accurate transmission toward task-oriented transmission. However, existing task-oriented image communication methods still face three major challenges: insufficient task-oriented Token representation, inadequate collaboration between Visual Tokens and Task Tokens, and limited interpretability of task decisions. To address these challenges, we propose an Explainable Task-Oriented Token Communication (ET-TokenCom) framework. By treating Tokens as unified units for information representation and transmission, the proposed framework constructs an end-to-end communication link that spans visual perception, wireless transmission, and task reasoning. At the transmitter, the ET-TokenCom framework extracts Visual Tokens from images to preserve low-level visual information. Meanwhile, Task Tokens generated by the FM are introduced to represent the target information and decision intent required by the current task. A Cross-Modal Attention (CMA) fusion mechanism is further designed, enabling Task Tokens to explicitly guide the selection, weighting, and transmission of Visual Tokens. At the receiver, the framework integrates Token decoding with an explainable output mechanism, where attention heatmaps are generated to highlight critical perceptual regions under different task objectives and reveal the influence of Task Tokens on the outputs. Finally, simulation results validate the effectiveness and robustness of the proposed ET-TokenCom framework.

Multimodal Large Language Models (MLLMs) achieve strong vision-language reasoning, but long visual contexts enlarge the KV cache and increase decoding latency. Existing compression methods rely on observation window attention for stable token-importance estimation, yet this aggregation can dilute sparse visual evidence and discard answer-critical tokens under aggressive compression. Therefore, we identify last-query attention as a complementary source for recovering such evidence, but its answer-irrelevant signals can mislead retention. We propose BACON, a plug-and-play method that calibrates observation window attention with last-query evidence and suppresses isolated noise via intra-layer coherence and inter-layer persistence. Across diverse benchmarks, models, budgets, and compression methods, BACON improves multimodal KV compression by 7.5% on average under the most aggressive budget, with gains up to 30.9%. Our project page is available at https://ryu1ion.github.io/official_BACON/

Clinicians diagnose brain tumors by synthesizing patient symptoms, medical history, and quantitative imaging data from modalities such as MRI and CT scans into a unified clinical judgement. However, most deep learning models rely on MRI/CT images alone, failing to replicate the clinicians multimodal reasoning. We explore a two-branch multimodal network combining raw MRI scans with 91 extracted radiomic features (intensity, texture, shape, and boundary descriptors) to classify brain tumors into glioma, meningioma, pituitary, and no-tumor. A pre-trained CNN backbone encodes the image stream, whereas a dedicated MLP encodes the radiomic stream. Both streams are fused via concatenation, gated, or bidirectional cross-modal attention strategies. Across nine experimental runs on a balanced 7,200 image dataset, all multimodal configurations outperform unimodal baselines with gated fusion achieving the best accuracy of 96.13%.

Parkinson's disease (PD) is a progressive neurodegenerative disorder that frequently causes speech impairments associated with hypokinetic dysarthria. As speech production relies on the precise coordination of complex neuromuscular mechanisms, speech analysis has emerged as a promising non-invasive and cost-effective biomarker for early PD detection. Recent deep learning approaches have shown encouraging results; however, most existing methods rely on a single speech representation, potentially overlooking complementary pathological information encoded across different feature spaces. In this work, we propose a multi-branch deep learning framework for automatic PD detection from speech. Each recording is segmented into 5-second chunks and represented using three complementary modalities: Log-Mel spectrograms, MFCCs, and HuBERT embeddings extracted from raw waveforms. The spectrograms are processed using a pre-trained ResNet-18 encoder, MFCC sequences are modeled through a BiLSTM network, and raw speech is encoded using a pre-trained HuBERT model. To effectively integrate these heterogeneous representations, we introduce a context-guided cross-modal attention mechanism that dynamically weights temporal HuBERT embeddings according to the global acoustic context derived from the spectrogram and MFCC branches. Experiments conducted on the publicly available Spanish PC-GITA corpus under strict speaker-independent 5-fold cross-validation demonstrate the effectiveness of the proposed approach. The proposed architecture achieves an accuracy of 91.51%, an F1-score of 91.24%, and an AUC of 95.97%. Furthermore, ablation studies confirm the contribution of both the proposed context-guided cross-modal attention mechanism and the integration of complementary speech representations. These findings highlight the potential of heterogeneous speech modeling for robust and clinically reliable PD detection.

Multimodal Large Language Models (MLLMs) face a significant inference bottleneck due to the quadratic computational cost of self-attention over long visual token sequences. However, we identify a critical inefficiency in current architectures: Visual Attention Saturation. Our analysis reveals that visual tokens rapidly establish their spatial structure and intra-modal relationships in early layers, rendering visual-to-visual self-attention in deeper layers computationally redundant. Conversely, Feed-Forward Networks (FFNs) in these layers remain essential for projecting visual features into the evolving textual semantic space. Leveraging this insight, we present Visual-Skip (V-Skip), a training-free inference paradigm that decouples spatial interaction from semantic evolution. Rather than discarding tokens, V-Skip imposes block-wise structured sparsity by selectively bypassing saturated visual self-attention modules. Furthermore, recognizing that varying downstream tasks demand distinct reasoning depths, V-Skip employs a lightweight, few-shot calibration to dynamically route the task-optimal sparsity path. Extensive experiments demonstrate that V-Skip effectively bypasses redundant vision attention to achieve block-wise sparsity, maintaining a 94.16% to 100.31% performance retention across diverse MLLMs. Ultimately, we prove that to reason more effectively, models do not need to discard what they see -- they simply need to "look less" at the right depth.

Transformer-based multimodal models rely on attention mechanisms to integrate information across heterogeneous modalities. Despite their success, existing multimodal attention formulations compute their scores through collections of pairwise dot-product interactions or by concatenating all the modalities into the keys, even when multiple modalities should be jointly involved. As a consequence, current approaches either incur quadratic complexity in the number of modalities or fail to explicitly model interactions that depend on the joint configuration of multiple representations. In this work, we introduce the Volumetric Multimodal cross-Attention (VMA), a novel cross-attention mechanism in which attention scores are defined as a function of the joint geometry of a query and multiple modality-specific keys. VMA computes the volume spanned by query and key vectors across multiple modalities, capturing joint multimodal dependencies beyond pairwise similarity, enabling native modeling of any-order modality interactions. We integrate VMA into our novel multimodal transformer architecture, named GRAMformer, explicitly designed to integrate any number of modalities. We evaluate the proposed model on multimodal learning tasks, demonstrating improved effectiveness and efficiency.

While Multimodal Large Language Models (MLLMs) demonstrate remarkable proficiency on complex vision-language tasks, the mechanisms by which they extract query-relevant visual features from complex, noisy contexts remain opaque. In this paper, we present an in-depth interpretability study that uncovers a profound structural property within MLLMs: functional sparsity in cross-modal retrieval. Leveraging a token-level metric termed Retrieval Attention Mass (RAM), we identify and characterize a highly specialized subset of attention heads, referred to as Context-aware Retrieval (CoRe) heads. Across diverse visual domains and model scales, we observe a clear functional division: CoRe heads act as dedicated information extractors, while most other heads distribute attention over broader contextual regions. Causal interventions further demonstrate the necessity of these specialized heads. Ablating only the top 5% of CoRe heads causes significant degradation in multimodal reasoning performance, whereas ablating lower-ranked heads has minimal effect. Moreover, acceleration experiments validate the utility of CoRe heads, showing that leveraging this localized sparsity significantly accelerates inference while maintaining robust task performance. Our findings reveal a structural principle of functional sparsity within MLLMs, refining the current understanding of mechanistic interpretability and laying a theoretical foundation that can inspire future architecture design and model optimization.

Existing human attention (saliency) modeling methods persist as highly fragmented across modalities, scenes, and task formulations. Consequently, even with increasing model capacity and data scale, current models predominantly remain scene-dependent and task-specific, failing to practically generalize in real-world applications. To address the fundamental limitations, we present the Attend to Anything Model (AAM), a multi-modal foundation model that unifies attention modeling across various image, video, and audio-visual tasks and scenes. AAM reformulates attention as a cognitive entailment relationship organized in a general-to-specific hierarchy, implemented through language prompts with hierarchical embeddings in hyperbolic space. Furthermore, to unify static image and dynamic video attention, we adopt a fluid-dynamics perspective, formulating video-frame attention as a diffusive temporal evolution governed by the Fokker--Planck equation. Extensive experiments on 16 benchmarks demonstrate that AAM consistently outperforms state-of-the-art methods by an average of 6% across various scenarios, while achieving approximately a 4$\times$ speedup in video inference. Overall, these results demonstrate that AAM provides a principled foundation for future research on attention and saliency-related tasks. The dataset and code will be available at https://github.com/wz-zhao/Attend-to-Anything.

Multimodal spatial reasoning often relies on long chains of intermediate textual and visual thoughts, where accumulating visual tokens and dense cross-modal attention incur substantial computation and memory overhead. To address this challenge, we propose Spectral-Progressive Thought Flow (SpecFlow), a novel lightweight multimodal spatial reasoning framework that represents intermediate visual thoughts in a fixed-size discrete cosine space. By exploiting strong energy compaction, SpecFlow preserves global layout and relational structure while introducing high-frequency details only when increased spatial precision is required. To align visual state evolution with linguistic intent, classifier-free guidance enables autoregressive textual thoughts to steer flow-based updates of the visual workspace/state without expanding the context. As a result, SpecFlow maintains a bounded visual workspace whose updates depend only on the current visual state and accumulated textual trace, enabling long-horizon inference with stable latency and memory usage independent of reasoning depth. Empirical results show that SpecFlow achieves competitive or superior reasoning performance while reducing computation and KV cache costs by up to 2.1 times.

Video understanding is shifting from the offline paradigm -- taking a fully recorded video as input and producing a single answer after it ends -- toward real-time interaction, in which the model perceives new frames while still replying, revises its answer as new evidence appears, and remains silent when there is nothing to say. We present MOSS-Video-Preview to validate this paradigm. Our central claim is that perception must not be blocked by generation; its natural realization is a two-channel architecture. We argue that a cross-attention backbone is better suited to real-time vision-language fusion than the prevailing decoder-only design: visual features enter through a side channel rather than joining the autoregressive sequence, so perception and generation run on separate, non-blocking pathways -- reducing the frequency of visual processing and exposing a clean channel-wise interface for independent compression. We complement this with a data synthesis pipeline that converts dense captions into real-time understanding QA whose answers are revised to match what the model has perceived so far, and we specialize an offline model on these data to elicit real-time behavior. Our model trails the strong Qwen2.5-VL-7B baseline overall -- a gap we attribute primarily to data and scale rather than the architecture -- yet attains competitive offline video and multimodal understanding, remains robust on the spatial and fine-grained temporal reasoning central to real-time use, and acquires behaviors that offline models lack: continuous perception, answer revision, and timely silence. On a single H200 with 256 frames per video, it achieves about a 5x speedup in time to first token and 2.7x higher decoding throughput, with negligible degradation in offline ability. Our study of paradigm, architecture, and data outlines a viable path toward real-time video understanding.

Current end-to-end autonomous driving systems predominantly rely on frame-based sensors, which suffer from inherent perception latency and motion blur during highly dynamic encounters, specifically sudden pedestrian crossings. To address this critical safety vulnerability, we propose DeepIPCv3, a novel multi-modal autonomous navigation framework that synergizes the dense 3D spatial geometry of LiDAR point clouds with the microsecond-level asynchronous event streams of a Dynamic Vision Sensor (DVS). We introduce a Transformer-inspired cross-modal attention mechanism to dynamically correlate these distinct modalities, allowing the network to instantaneously prioritize high-speed dynamic updates without sacrificing structural scene awareness. The fused latent representations are then mapped to safe local waypoints and executable control commands via a hybrid policy network that blends heuristic trajectory tracking with direct neural predictions. Due to the severe physical risks associated with live testing of these sudden crossing scenarios, the framework is rigorously evaluated offline using a custom multi-modal dataset collected across both well-illuminated noon and challenging evening conditions. Extensive comparative and ablation studies demonstrate that DeepIPCv3 achieves state-of-the-art predictive performance. By effectively eliminating exposure failures and motion blur, the proposed LiDAR and DVS fusion yields the lowest trajectory and control command errors, enabling highly reactive, mathematically bounded evasive maneuvers regardless of ambient illumination. To support future research, we will release the codes to our GitHub repo at https://github.com/oskarnatan/DeepIPCv3.

Pre-trained video large language models excel at visual reasoning. However, they struggle when videos arrive with auxiliary streams, such as audio, depth map, or dense temporal evidence. In such a scenario, uniform fusion induces modality interference, allowing irrelevant channels to distract the model. To address this issue, we present a unified multimodal video understanding framework, named UniMVU, that performs instruction-aware fusion across video, audio, depth map, or any other modality inputs via two levels of dynamic gating: inner-modality gates emphasize salient regions within each modality, whereas modality-level gates re-weight whole streams; both are conditioned on the text instruction to adaptively balance modality importance. Our UniMVU combines cross-modal self-attention with instruction-driven inner-modality gating module and a modality-level gating module with control token; for time-aligned streams we further adopt a fast-to-slow fusion scheme that reduces redundancy. Across six benchmarks (AVQA, AVSD, Music-AVQA, ScanQA, SQA3D and MVBench), our UniMVU achieves consistent gains over static-fusion baselines achieving gains as high as 13.5 in terms of CIDEr metric. Further, our analysis shows that the gating mechanism aligns with the human-interpretable modality relevance, and ablations show the contributions of inner-modality and modality-level gating. Our UniMVU provides a simple, unified recipe for instruction-aware multimodal video understanding that scales to diverse modalities without hand-crafted fusion rules.

Recent feed-forward models have significantly advanced geometry perception for inferring dense 3D structure from sensor observations. However, its essential capabilities remain fragmented across multiple incompatible paradigms, including online perception, offline reconstruction, multi-modal integration, long-horizon scalability, and metric-scale estimation. We present UniT, a unified model built upon a novel Group Autoregressive Transformer, which reformulates these seemingly disparate capabilities within a single framework. The key idea is to treat groups of sensor observations as the basic autoregressive units and predict the corresponding point maps in an anchor-free and scale-adaptive manner. More specifically, diverse view configurations in both online and offline settings are naturally unified within a single group autoregression process. By varying the group size, online mode operates over multiple autoregressive steps with single-frame groups, whereas offline mode aggregates a multi-frame group in a single forward pass. Meanwhile, a queue-style KV caching mechanism ensures bounded autoregressive memory over long horizons. This is enabled by reducing long-range dependencies on early frames through anchor-free relational modeling, thereby allowing outdated memory to be discarded on the fly. To improve metric-scale generalization across scenes, a scale-adaptive geometry loss is further introduced within this framework. It couples relative geometric constraints with a partial absolute scale term, implicitly regularizing global scale and inducing a progressive transition from scale-invariant geometry to metric-scale solutions. Together with a dedicated modal attention module for integrating auxiliary modalities, UniT achieves state-of-the-art performance in unified geometry perception, as validated on ten benchmarks spanning seven representative tasks.

Large multimodal models (LMMs) inherit the self-attention mechanism of pretrained language backbones, yet standard attention can exhibit suboptimal allocation, including cross-modal misallocation between textual and visual evidence and intra-visual imbalance among visual tokens. We propose RAVE (Re-Allocating Visual Attention), a lightweight pair-gating mechanism that adds a learned query-key bias to pre-softmax attention scores over visual keys, derived from pre-RoPE query and key features. RAVE requires no architectural modification to the backbone and can be trained end-to-end with the rest of the model. Across a suite of multimodal benchmarks, RAVE improves over standard attention by an average of 3 points, with the largest gains on perception-intensive tasks -- including multilingual OCR, chart understanding, document VQA, and scene text VQA -- where accurate visual grounding is critical.

We present SWIM (See What I Mean), a novel training strategy that aligns vision and language representations to enable fine-grained object understanding solely from textual prompts. Unlike existing approaches that require explicit visual prompts, such as masks or points, SWIM leverages mask supervision only during training to guide cross-modal attention, allowing the model to automatically attend to the user-specified object at inference. Our cross-attention analysis of pretrained multimodal large languagemodels (MLLMs) reveals a systematic discrepancy: Attribute words produce sharp, localized activations in the visual modality, whereas object nouns yield diffuse and scattered patterns due to semantic reference bias and distributed high-level representations. To address this misalignment, we construct NL-Refer, an enriched dataset, in which each object mask is paired with a precise natural language referring expression. SWIM extracts multi-layer cross-attention maps from object nouns and enforces spatial consistency with ground-truth masks. Experimental results demonstrate that SWIM substantially improves text-visual alignment and achieves superior performance over visual-prompt-based methods on fine-grained object understanding benchmarks. The code and data are available at \href{https://github.com/HumanMLLM/SWIM}{https://github.com/HumanMLLM/SWIM}.

Large vision-language models (LVLMs) have emerged as a powerful paradigm for multimodal intelligence, but their growing deployment also expands the attack surface of prompt injection. Despite this growing concern, existing attacks still suffer from a critical limitation: the injected prompt for one modality only steers the model's interpretation of that singular input. Alternatively, these attacks remain multimodal but fail to achieve cross-modal prompt perturbation. To bridge this gap, we introduce a novel cross-modal prompt injection attack CrossMPI, which can steer the model's interpretation of both textual and visual inputs via image-only prompt injection. Our design is underpinned by the following key breakthroughs. First, we turn the focus of the injected prompt perturbation optimization from the visual embedding space (typically with only $10^5$ parameters) to the model hidden state space (for multimodal information integration and with $10^7$ parameters). Then, two strategies are adopted to mitigate the optimization challenges posed by the larger parameter space. To constrain the optimized model parameter space, we introduce a layer selection strategy that identifies the layers most critical to multimodal integration. Interestingly, deviating from the past experience, our analysis reveals that the optimal layers for LVLM prompt perturbation reside in the middle of the model rather than the last. To constrain the image perturbation space, we propose a new distance-decremental perturbation budget assignment strategy that allocates budgets decrementally as the pixel distance to semantic-critical regions increases. Extensive experiments across multiple LVLMs and datasets show that our method significantly outperforms baseline approaches.

Recent diffusion-based methods for material transfer rely on image fine-tuning or complex architectures with assistive networks, but face challenges including text dependency, extra computational costs, and feature misalignment. To address these limitations, we propose MaTe, a streamlined diffusion framework that eliminates textual guidance and reference networks. MaTe integrates input images at the token level, enabling unified processing via multi-modal attention in a shared latent space. This design removes the need for additional adapters, ControlNet, inversion sampling, or model fine-tuning. Extensive experiments demonstrate that MaTe achieves high-quality material generation under a zero-shot, training-free paradigm. It outperforms state-of-the-art methods in both visual quality and efficiency while preserving precise detail alignment, significantly simplifying inference prerequisites.

Large vision-language models (LVLMs) have achieved remarkable performance across diverse multimodal tasks, yet they continue to suffer from hallucinations, generating content that is inconsistent with the visual input. Prior work DHCP (Detecting Hallucinations by Cross-modal Attention Pattern) has explored hallucination detection from the perspective of cross-modal attention, but does not address hallucination mitigation. In this paper, we propose MHSA (Mitigating Hallucinations via Steered Attention), a lightweight framework that mitigates hallucinations by learning to correct cross-modal attention patterns in LVLMs. MHSA trains a simple three-layer MLP generator to produce corrected attention, guided by supervisory signals from the DHCP discriminator and the LVLM itself. During inference, MHSA mitigates both discriminative and generative hallucinations across various datasets and LVLMs by simply replacing the original cross-modal attention with the corrected one, without modifying any LVLM parameters. By extending cross-modal attention mechanisms from hallucination detection to hallucination mitigation, MHSA offers a novel perspective on hallucination research in LVLMs and helps enhance their reliability.

Multimodal Large Language Models (MLLMs) have shown strong performance in multi-image cross-modal retrieval, yet suffer from severe position bias, where predictions are dominated by input order rather than semantic relevance. Through empirical analysis, we identify a phenomenon termed Logit-Attention Divergence, in which output logits are heavily biased while internal attention maps remain well-aligned with relevant visual evidence. This observation reveals a fundamental limitation of existing logit-level calibration methods such as PriDe. Based on this insight, we propose a training-free, attention-guided debiasing framework that leverages intrinsic attention signals for instance-level correction at inference time, requiring only a minimal calibration set with negligible computational overhead. Experiments on MS-COCO-based benchmarks show that our method substantially improves permutation invariance and achieves state-of-the-art performance, enhancing accuracy by over 40% compared to baselines. Code is available at https://github.com/brightXian/LAD.

Audio-Visual Segmentation (AVS) targets pixel level localization of sounding emitting objects in videos. However, existing models rely on dense cross-modal attention with quadratic computational cost, limiting their suitability for resource efficient deployment. Most efficiency oriented methods focus on backbone reduction and overlook the interaction module as the primary bottleneck. This paper proposes LightAVSeg, a lightweight framework that replaces heavy attention with a decoupled design for semantic filtering and spatial grounding, resulting in interaction costs that scale linearly with spatial resolution. Furthermore, we introduce an auxiliary alignment loss to enforce semantic consistency during training with zero inference overhead. Extensive experiments demonstrate that LightAVSeg achieves a new state-of-the-art among lightweight methods: with 20.5M parameters ~1/7 of AVSegFormer), it reaches 50.4 mIoU on the MS3 benchmark and enables efficient inference on a mobile processor.

Pathological complete response (pCR) is a key prognostic factor in breast cancer patients undergoing neoadjuvant therapy, strongly associated with long-term survival and treatment personalization. However, accurate pre-treatment pCR prediction remains challenging due to severe class imbalance and limited generalizability across diverse clinical settings. In this work, we propose a multimodal stepwise clinically-guided attention learning framework for pCR prediction from breast magnetic resonance imaging (MRI), designed to address these limitations through medically grounded spatial guidance and multimodal integration. The approach follows a stepwise training strategy inspired by physician reasoning: the model first learns global discriminative imaging patterns, then attention mechanisms are introduced to constrain the network toward tumor regions, and finally clinical variables are integrated to refine decision-making. This guidance strategy encourages prioritization of task-relevant features, improving identification of responders despite their limited representation in the dataset. Moreover, grounding attention in anatomically consistent tumor regions reduces reliance on dataset-specific patterns, thereby enhancing cross-institutional generalization. The framework is evaluated through external validation across heterogeneous MRI cohorts. Compared to non-guided single-stage baselines, the proposed approach improves sensitivity while maintaining competitive specificity, and produces anatomically coherent attention maps that support interpretation of the model's predictions. These findings highlight the potential of clinically-guided multimodal attention learning for robust and generalizable pCR prediction in breast cancer.

While Vision-Language-Action models (VLAs) have shown remarkable progress toward human-like generalist robotic policies through the versatile intelligence (i.e. broad scene understanding and language-conditioned generalization) inherited from pre-trained Vision-Language Models, they still struggle with complex real-world tasks requiring broader functional capabilities (e.g. motion awareness, long-term memory, and physical sensing). To address this, we introduce RLDX-1, a general-purpose robotic policy for dexterous manipulation built on the Multi-Stream Action Transformer (MSAT), an architecture that unifies these capabilities by integrating heterogeneous modalities through modality-specific streams with cross-modal joint self-attention. RLDX-1 further combines this architecture with system-level design choices, including data synthesis for rare manipulation scenarios, learning procedures specialized for human-like manipulation, and inference optimizations for real-time deployment. Through empirical evaluation, we show that RLDX-1 consistently outperforms recent frontier VLAs (e.g. $π_{0.5}$ and GR00T N1.6) across both simulation benchmarks and real-world tasks that require broad functional capabilities beyond general versatility. In particular, RLDX-1 shows superiority in ALLEX humanoid tasks by achieving success rates of 86.8% while $π_{0.5}$ and GR00T N1.6 achieve around 40%, highlighting the ability of RLDX-1 to control a high-DoF humanoid robot under diverse functional demands. Together, these results position RLDX-1 as a promising step toward reliable VLAs for complex, contact-rich, and dynamic real-world dexterous manipulation.