Cross-Attention

Synthetic image and mask pairs can alleviate scarce colonoscopy annotations, but realistic synthesis requires preserving the supplied lesion while generating compatible mucosa. Existing foreground-guided methods treat all non-foreground pixels as background and rely mainly on local integration. Directly applying them to colonoscopy causes two problems: non-mucosal black regions contaminate generated tissue, and local reasoning produces inconsistent mucosal texture and illumination. We propose LAMP, the first foreground-guided framework for polyp image synthesis based on lesion-guided adaptive mucosal context propagation. LAMP explicitly separates the lesion, valid mucosa, and camera exterior using a field-of-view (FOV) mask. Lesion-to-Mucosa cross-attention extracts lesion appearance conditions for valid-mucosa locations, while FOV-constrained multidirectional Vision Receptance Weighted Key Value propagates them over legal tissue support. An adaptive gate then controls their residual fusion into the diffusion U-Net. Extensive experiments on five polyp datasets demonstrate that LAMP substantially outperforms existing methods in overall generation quality and consistently improves five downstream segmentation models. Our code will be released at https://github.com/wangtong627/LAMP.

Effective EEG decoding requires representations that preserve organization among channels, local waveform dynamics, and long-range temporal context. Existing EEG architectures often capture these structures using separate specialized modules or collapse them into a single token sequence, making it difficult to maintain their distinct roles and coordinate their interactions throughout the backbone. We propose TriDim, a reusable block that preserves the representation shape and keeps three EEG axes explicit: channel, sample position within each patch, and patch position across the recording. These axes correspond to spatial, short-term temporal, and long-term temporal information, respectively. Each TriDim block applies feed-forward transformations along individual axes and cross-axis attention to coordinate information exchange among them. By stacking TriDim blocks with a multi-level tri-axis readout, we construct TriDimEEG, a standalone EEG decoder. Under strict cross-subject evaluation on eight datasets spanning clinical diagnosis, sleep staging, motor imagery, and emotion recognition, TriDimEEG achieves the best overall performance among fifteen evaluated models, with a 4.3% relative improvement in average accuracy over the second-best model. Replacing Transformer blocks in three EEG foundation models with TriDim blocks yields an average relative improvement of 7.4% in downstream accuracy while reducing parameter counts by 17.0% to 47.3%. These results establish TriDim as an effective and reusable building block and TriDimEEG as a strong standalone EEG decoder. Code and parameters of TriDimEEG are available at https://github.com/ncclab-sustech/TriDim_model.

We introduce a simple architectural modification to decoder-only transformers: a persistent recurrent state that observes hidden representations via cross-attention, updates itself through a GRU, and modulates subsequent processing via gated addition. Inserted between the lower and upper halves of a 6-layer transformer, this module adds only 3.7% additional parameters while reducing evaluation loss from $2.438 \pm 0.004$ to $1.743 \pm 0.018$, corresponding to a 28.5% reduction on held-out language modeling data. The improvement is statistically significant across 5 random seeds ($p < 0.01$) and corresponds to reduced overfitting (generalization gap 0.12 vs 0.26). Through controlled ablations, we demonstrate that the improvement stems entirely from the persistent memory topology, not from auxiliary self-prediction objectives. A model with identical topology but no auxiliary loss performs equivalently, while a random auxiliary loss provides no benefit. Representation probing reveals that the persistent state encodes narrative position (52% vs 33% chance level)---information that standard attention maintains less efficiently. Our results suggest that bridging transformer layers with a lightweight recurrent memory is a simple, effective approach to improving generalization in small-scale language models.

Large-scale vision-language models (VLM) such as CLIP enable strong open-vocabulary reasoning, yet deploying these capabilities on resource-constrained edge devices remains challenging. EdgeVL addresses this problem by distilling CLIP representations into lightweight multi-modal encoders and applying quantization-aware training (QAT) for efficient Open-Vocabulary Classification (OVC) on edge hardware. However, its two-stage optimization applies different objectives for distillation and QAT, and contrastive learning is performed within the quantized student space, which can result in inconsistent optimization and reduced training efficiency. Moreover, identical supervision across RGB and non-RGB modalities may lead to modality imbalance. We propose a unified framework for quantized semantic distillation tailored to edge deployment. By jointly optimizing distillation and quantization within a unified teacher-anchored framework, our method ensures consistent training under quantization, suppressing hard negatives and enlarging decision margins. Additionally, we design a lightweight cross-attention adapter that enhances non-RGB representations through RGB-guided semantic transfer, narrowing the modality gap. Extensive experiments demonstrate consistent improvements on non-RGB modalities while maintaining deployment efficiency.

Humanoid robots often execute motion commands through whole-body controllers (WBCs) that track targets while maintaining balance and stability. However, most WBCs are blind to scene geometry, which can lead to collisions from imperfect target motions that are geometrically unsafe due to perception, planning, or teleoperation errors. We propose RECAL, a Robot--Environment Cross-Attention Layer that wraps a blind WBC to trade off target tracking against collision avoidance using external scene geometry. RECAL supports collision-aware tracking of floating-base and end-effector commands, including collision avoidance for held objects. It represents the robot, held objects, and environment as point clouds, using cross-attention between robot/object points and the environment to produce geometry-aware control features. In simulation, RECAL improves collision avoidance while preserving target-tracking performance across frozen-arm and adaptive-arm locomotion, object-carrying, and standing-manipulation scenarios relative to alternative geometry-aware WBC architectures. We further demonstrate the controller on a real Digit V3 humanoid robot.

Efficient microservice scheduling is crucial for maintaining load balance across nodes in data centers and ensuring high quality of service. However, achieving this in practice remains challenging due to dynamic resource imbalance under fluctuating workloads, nonlinear coupling across multiple resource dimensions, and the heterogeneity of microservice resource demands. While reinforcement learning-based approaches have shown promise, they struggle to capture the complex interdependencies among heterogeneous resources and neglect the importance of learning informative system representations. To address these limitations, we propose MCRL2, a novel reinforcement learning approach augmented with multi-resource cross-attention-based representation learning for microservice scheduling. Specifically, we first propose MCRL, a novel representation learning approach that captures structured and informative interactions among nodes, resources, and microservices via a multi-resource cross-attention mechanism. Then, MCRL2 augments reinforcement learning through MCRL-enhanced actor-critic architecture combined with a maximum entropy objective, improving system state expressiveness and leading to more stable and effective scheduling decisions. Extensive experiments on real production cluster traces demonstrate that MCRL2 significantly outperforms existing baselines in load balancing, scheduling success rate and average completion time across diverse workload patterns.

Driver gaze provides information regarding driver visual attention and situational awareness to the surrounding traffic. Existing driver gaze estimation studies represent gaze in terms of gaze zone or gaze vector/point-of-gaze (PoG). However, object-level gaze information provides a more semantically meaningful representation of visual attention by identifying attended objects, such as vehicles, pedestrians, or traffic signals. In this study, we propose an end-to-end driver gaze object prediction framework, TransGaze-Object, Transformer-based Gaze Object prediction model. The proposed framework first extracts facial features, including face and iris-weighted eye features, along with trafficobject spatial features. A transformer based cross-attention mechanism is then used to compute similarity scores and attention weights for predicting the drivers gaze object. To train this model, we propose a benchmark driver gaze dataset, Urban Driving-Face Scene Gaze (UD-FSG), comprising synchronized driver-face and traffic-scene images, scene objects bounding boxes, and gaze labels in terms of 2D gaze coordinate and gaze object. The TransGaze-Object model achieves an overall accuracy of 60% for gaze-object prediction, compared to 51% accuracy obtained from associating the estimated Point-of-Gaze to traffic objects. The error analysis reveals that TransGaze-Object reduces confusion between traffic objects (predicted) and the background (ground-truth), achieving an error rate of 11.68%, a 49.7% relative reduction compared with 23.21% error obtained from PoG-based gaze-object association. Overall, the results demonstrate the effectiveness of directly predicting gaze objects from driver-face and traffic-scene information, rather than estimating an intermediate Point-of-Gaze and subsequently associating it with traffic objects.

Modern vision-language-action (VLA) policies predict a whole chunk of actions: one to two seconds of coordinated motion emitted in a single forward pass. Yet an action chunk is essentially a short multivariate trajectory, but inside these models it is a sequence of generic per-timestep hidden tokens decoded by a linear head. This under-serves two motion structures. First, frequency: a chunk superimposes a smooth global trend and fine corrective motion across time scales, and a single token entangles them. Second, cross-phase geometry: motions of different phases (reach, contact, grasp adjustment, settling) unfold along very different, near-orthogonal directions in representation space, yet are tightly related for the task and arise across the time axis. Dot-product attention scores alignment by an inner product, so it favors aligned tokens and is least sensitive near orthogonality, leaving such relationships for the network to recover through a detour. We introduce Time-Frequency Geometric Cross-Attention (TFGCA), a drop-in module repairing both blind spots. TFGCA uses a per-dimension learnable stationary wavelet transform to decompose the action chunk into time-frequency tokens, and each time token retrieves information from them via a cross-attention that fuses the dot product (similarity) with the wedge-product magnitude (sensitive to near-orthogonality) through a learnable weight. A zero-initialized residual reproduces the base behavior at initialization, so it can be dropped onto a pretrained VLA and fine-tuned jointly. Relative to the same-source base, TFGCA improves in-distribution LIBERO by +1.5 on average, the OOD LIBERO-Plus by +6.3, the randomized average under RoboTwin domain randomization by +28.5, and the overall success rate on three real-robot AgiBot A2 tasks by +11.67 points, with larger gains out of distribution.

Medical claims data comprise the financial details, including the expenses and billing information, as well as the clinical information, such as the diagnoses and treatments, of patients visiting medical facilities. Recently, it has been acknowledged that large databases can be constructed from medical claims data for medical research purposes. However, the clinical information within these datasets is often medically unstructured, limiting its application in comprehensive analyses. This study enhances predictive model performance for major adverse cardiovascular events (MACE), a leading cause of death worldwide. Models that predict MACE are crucial to clinical practice guidelines. We utilize a cross-attention mechanism to develop a method that effectively weights the relationships between diagnoses and treatments. Effectively repre- senting the clinical information contained in medical claims data, this approach generates more representative features for predicting MACE. The ROC-AUC score of our proposed cross-attention-based model was 0.7720, higher than other benchmark models including the conventional atherosclerotic cardiovascular disease model, the light gradient boosting machine, and a self-attention-based model. These results indicate that integrating the clinical structure of medical claims data using a cross-attention mechanism significantly enhances the performance of predictive models.

Modeling long-term user behavior is central to sequential recommendation and billion-scale industrial recommender systems, yet production ranking models operate under strict latency, memory, communication, and training-throughput constraints. At the 100K scale, the challenge extends beyond attention complexity: raw sequence features must be stored, transferred, and repeatedly processed during training and online serving. Existing approaches based on history truncation, multi-stage behavior retrieval, compressed lifelong histories, or train-short/infer-long extrapolation either weaken end-to-end optimization or retain substantial length-dependent cost. We present SequenceO1, an end-to-end framework for ultra-long user behavior sequence modeling, deployed at full traffic on Douyin with histories of up to 100K interactions. SequenceO1 follows a compress-then-reason design. Its Sketch Attention (SA) uses learnable prototypes and prototype-wise normalization to compress the raw history into a fixed-size, target-agnostic user representation. Target-conditioned Stacked Target-to-History Cross Attention (STCA) then models complementary time scales: a recent 10K suffix for short-term interests and the compact sketch for long-term preferences. To make training and inference practical, SequenceO1 combines low-rank user representation caching, multi-request user-level batching, pipeline lift, and a fused FlashSA kernel to amortize feature storage, communication, and computation across targets, training instances, and consecutive requests. Production experiments show consistent offline and online gains, while the compact cached sketch retains most of the benefit of directly scaling end-to-end sequence ranking to 100K. These results provide a practical model-system approach to efficient attention, sequence compression, and scalable long-sequence and long-context recommendation systems.

Infrared small target detection (ISTD) is an important research direction in image processing. However, existing methods are limited by severe background noise propagation and target degradation in high-level semantic features. To address these limitations, this paper proposes a plug-and-play physics-informed cascaded asymmetric network, named PICANet. Specifically, we construct a hierarchical prior decoupling module to explicitly extract low-level and high-level physical information, thereby characterizing target features at different levels rather than relying solely on convolutional extraction. Furthermore, a dual-prior interactive fusion module is developed to dynamically refine target representations while suppressing complex background clutter. Unlike previous work, a multi-level cross-feature attention module with the cascaded asymmetric mechanism is introduced to achieve precise alignment between high-level semantics and low-level spatial details. Extensive experiments demonstrate that the proposed PICANet outperforms state-of-the-art ISTD methods, showing satisfactory detection accuracy even against complex backgrounds. Our code is available at https://github.com/xianchaoxiu/PICANet.

Pairwise preference labels rank complete images, yet Diffusion-DPO applies their effect over many spatial and denoising-time coordinates. For attention-based, noise-prediction latent diffusion, ToPO (Token-Oriented Preference Optimization) constructs a per-minibatch, detached, separable spatial-temporal route from branchwise squared-residual contrast in a frozen reference denoiser. Preferred-branch cross-attention uses content tokens to modulate the spatial factor, and an auxiliary pixel-midpoint ordering term is added without local labels or a learned reward model. In matched three-seed retrainings with a shared update schedule, ToPO has higher endpoint estimates than Diffusion-DPO on all five reported SD-1.5 metrics and on HPSv2, ImageReward, and CLIP for SDXL. It also receives larger raw win shares in an aggregate blind SDXL A/B study. These findings are scoped to the reported equal-update U-Net protocols rather than an equal-compute comparison.

Action chunking is a standard execution strategy in modern Vision-Language-Action (VLA) frameworks, but fixed execution horizons impose a trade-off between efficiency and accuracy. Short chunks require frequent inference and may cause oscillatory behavior, whereas long chunks can become misaligned with newly observed states. We address this limitation with an adaptive action chunking approach based on internal cross-attention dynamics in the action expert. We observe that, as the prediction horizon extends, action-to-observation cross-attention becomes increasingly dispersed and its entropy rises toward a plateau. This pattern is associated with higher action prediction error and provides an online signal that the current observation offers limited grounding for further open-loop execution. Based on this observation, we introduce a training-free truncation mechanism that detects sustained high-entropy plateaus and dynamically selects the execution horizon during inference. The method uses attention weights already computed by the policy and introduces negligible additional overhead. Evaluations on $π_{0.5}$ and X-VLA across RoboTwin 2.0, LIBERO, and three real-world manipulation tasks show improved average task success over fixed-horizon and adaptive chunking baselines, while preserving efficient closed-loop control. These results show that cross-attention dynamics can provide a practical internal signal for adaptive action execution in VLAs.

Speech enhancement is often used as a front-end for robust ASR, yet recurrent temporal and cross-band modules introduce sequential dependencies that reduce parallel efficiency. In this paper, we present a sequence-parallel band-split enhancement front-end built on a Parallel Time-Band Mixer (PTBM) block that eliminates within-block recurrent unrolling. PTBM integrates intra-band temporal mixing and per-frame cross-band attention within a unified parallel architecture, enabling efficient contextual modeling across both time and frequency dimensions. The system retains the mask-plus-residual reconstruction interface and introduces learned Observation-Adding (LOA) to suppress ASR-sensitive artifacts without development-set tuning. Experiments on DNS Challenge and CHiME-4 with frozen Whisper back-ends show that the proposed front-end consistently reduces word error rate relative to recurrent band-split baselines while requiring only 0.96 M parameters and 0.58 GMAC/s for the front-end network.

Multimodal information systems increasingly route generated visual content back through the same vision-language index that informed its production, so the output must remain retrievable by the queries it was meant to serve. When the scene contains entities at vastly different scales, existing language-guided generators condition on a single, globally pooled text embedding and quietly drop scale-specific concepts, breaking concept-query retrieval even when pixel fidelity is high. We formalise this failure as semantic collapse and propose CERES, a closed-loop multimodal indexing framework that builds a three-level semantic pyramid, mines implicit concepts via a co-occurrence-aware router, performs scale-routed cross-attention into a lightweight U-Net generator, and verifies coverage by re-indexing the generated image with the same frozen VLM. A continuously differentiable soft-Jaccard coverage objective returns dense gradients to the 0.39 M-parameter generator under explicit non-degeneracy conditions, and coverage is verified by an independent DINOv2 linear probe trained only on external scene and object labels. On four pansharpening benchmarks across seven settings, CERES delivers the new state of the art with the largest gains where scale variation is most extreme (+4.64% relative Q2n and +9.7 mAP for DOTA detection). It also improves concept-query retrieval Recall@5 by +14.0 points and image-text mean reciprocal rank by 0.19 over the strongest baseline, showing that the closed loop preserves queryable content rather than self-referential feature consistency.

Fine-grained human action recognition (FHAR) must distinguish visually similar actions that differ mainly in body configuration, timing, or local appearance. RGB representations retain visual context but often suppress joint-level geometry, whereas skeleton representations encode kinematics but discard dense spatial detail. We introduce FineX, which factorizes fine-grained cues into RGB appearance, pose heatmap geometry, and skeletal-graph topology. Pairwise cross-attention enables symmetric, stream-preserving information exchange, followed by a streamwise latent sparse Mixture-of-Experts that routes each representation to a content-dependent subset of shared experts, regularized by a load-balancing objective. FineX achieves state-of-the-art results on Gym99, Gym288, and Diving48. On the long-tailed Gym288, it raises mean class accuracy from 68.6% to 76.2% (+7.6 points) without textual supervision or large-scale vision-language pre-training, demonstrating the benefit of structured visual-pose-graph fusion and conditional expert refinement for FHAR.

Text-to-Video (T2V) generative models are vulnerable to jailbreak attacks in real-world deployment, leading them to produce harmful or inappropriate content. Existing defense approaches mainly rely on input filtering or reconstruction, which not only incur high computational latency but also tend to distort semantics. To address these issues, we experimentally and systematically analyze the differences between clean and jailbreak samples in the cross-attention feature space, revealing for the first time a cumulative separation effect and a progressively increasing trend of linear separability between the two during the diffusion process. Based on this insight, we propose SafeCA, a feature-level defense mechanism for safe cross-attention localization and regularization. Firstly, we identify key defensive regions and values through attention stability analysis using cross-attention features collected from clean prompts within a single inference. Secondly, SafeCA mitigates anomalous activations via attention masking with energy normalization and introduces a lightweight semantic-space adapter to redirect abnormal semantic flows. Furthermore, we detect and suppress potentially malicious tokens by back-propagating feature anomaly signals to the input cue words, thereby enhancing the deployability of the defense in commercial models. Experimental results show that SafeCA reduces the jailbreak success rate by about 20% on mainstream T2V models, adds almost no inference overhead (+0.1s), and maintains good text-video semantic consistency. Overall, SafeCA provides an architecture-level, deployable protection paradigm for T2V generation models.

Multi-oriented text is ubiquitous in real-world scenes and remains a major challenge for scene text recognition (STR). Existing rotation-aware methods explicitly estimate text orientation. However, due to the lack of theoretical guarantees, they are prone to error accumulation, increased computational cost, and strong reliance on data. In this work, we incorporate rotation invariance into the STR framework to address these limitations. Specifically, we adopt an encoder-decoder architecture, embedding rotation equivariance in the encoder and rotation invariance in the decoder to construct a fully rotation-invariant network. On the decoder side, we first identify and prove the rotation-invariant property of the cross-attention mechanism and use it to formulate a rotation-invariant text decoder that maps visual features to output text in a rotation-invariant manner. On the encoder side, we propose a rotation-equivariant local-global extraction network that integrates deep equivariant convolutions with self-attention, enabling rotation-equivariant feature extraction while modeling inter-character dependencies and preserving fine-grained visual details. By integrating the encoder and decoder, we obtain an end-to-end Rotation-Invariant Scene Text Recognition network (RISTER). RISTER provides rotation invariance with theoretical guarantees, enhancing robustness on multi-oriented samples without introducing additional inference computation or relying on data-driven orientation correction. Experiments show that RISTER achieves state-of-the-art performance on both standard and multi-oriented benchmarks, surpassing the second-best model by 4.0 percent in accuracy on the general multi-oriented dataset.

Transformer layers generate state-dependent interaction networks: token representations determine the attention matrix, which in turn updates the representations. We study this feedback in a minimal normalized self-attention dynamics and identify the overlap gap as the central quantity governing its attractor structure in the thermodynamic limit. When tokens form internally aligned clusters and their similarity to members of the same cluster exceeds that to every other cluster by a nonvanishing amount, inter-cluster attention is exponentially suppressed as the dimension increases. This mechanism produces a high-dimensional manifold of clustered fixed points, ranging from a few macroscopic clusters to extensive microscopic fragmentation, and also controls their stability against perturbations. Starting from an unstructured Gaussian state, we find that clustered states nucleate from the diffuse background only above a finite threshold in attention sharpness, giving rise to a dynamical attention-condensation transition.

Multispectral object detection combines visible and thermal imagery to improve perception under challenging illumination and environmental conditions. However, differences in modality appearance and reliability can introduce redundant or conflicting responses, limiting the use of complementary information. Complex fusion mechanisms further increase computational cost, creating a persistent trade-off between detection accuracy and efficiency. To address these challenges, CFGPNet is proposed, a cross-attention-based fused gradient programmed network. The framework incorporates re-parameterized RepViT blocks into the YOLOv9 architecture to strengthen spatial and channel representations while maintaining efficient feature extraction. Cross Computation Efficient Attention (CrossCEA) exchanges spatial attention maps between modalities at multiple detection scales, allowing each stream to emphasize regions supported by the other while preserving modality-specific information. Attention Selection and Aggregation Fusion (ASAF) combines dense feature aggregation with selection of the strongest responses from multiple attention branches to form compact, discriminative fused representations. A programmable gradient information pathway provides auxiliary supervision during training to improve feature learning. This pathway is removed after training, adding no parameters or operations at inference. Experiments on FLIR, M3FD, LLVIP, VEDAI, and MFAD demonstrate favorable accuracy-efficiency trade-offs across three model scales, with the smallest variant requiring 15.3 million parameters and 56.9 GFLOPs. The code is available at https://github.com/NimaHatami99/CFGPNet.

In this paper, we present TRACE (Tokenized Robust Attention for Contact-Aware Estimation), an end-to-end learned proprioceptive odometry estimator for legged robots under unreliable contact conditions. The proposed estimator directly predicts relative displacement, relative rotation, and body-frame velocity from a recent history of onboard inertial and joint measurements. To improve robustness under unreliable contact conditions, we introduce a foot-aware cross-attention module that adaptively weights IMU and leg-wise kinematic tokens without relying on manually defined contact or slip thresholds. The estimator is trained with direct supervision and two physics-inspired auxiliary losses that promote kinematic consistency and reliable use of leg information. To reduce policy-specific overfitting and consequently improve sim-to-real transfer, simulation training incorporates policy randomization, followed by partial real-world fine-tuning of the temporal encoder and prediction head. Experiments across diverse indoor and outdoor terrains demonstrate consistent reductions in position drift compared with classical filtering-based, hybrid, and purely learning-based baselines. Ablation studies further validate the contributions of the proposed training objectives, policy randomization, and real-world fine-tuning, particularly under unreliable contacts and sim-to-real mismatch.

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.

Modern recommendation models gain prediction quality by scaling feature-interaction and sequence modules, but production cost constraints cap how far systems can scale. In this work, we propose Request-Oriented Compute Sharing (ROCS), a modeling and inference paradigm that exploits a unique property of recommendation inference: each user request is evaluated against many candidates, while request-side features are shared across candidates. ROCS defers request-candidate interactions as late as possible, isolates candidate-dependent representations, and evaluates substantial portions of the model once per request rather than once per candidate, significantly improving inference efficiency while maintaining or improving prediction quality. To realize this paradigm, we develop Generalized Layer Masking (GLM) to enforce candidate isolation in feature-interaction architectures, and Deep Cross Attention (DCA) to extend request-oriented sharing to sequence architectures. To support efficient GPU deployment, we co-design In-Kernel Broadcast Optimization (IKBO) that significantly accelerates ROCS model execution. Experiments on public benchmarks show that ROCS consistently improves the quality-efficiency tradeoff across recommendation backbones. On production-scale workloads, ROCS achieves up to a 3x QPS improvement on retrieval models without quality degradation and a 0.5% relative LogLoss improvement with a 50% QPS gain on a short-form video ranking model. ROCS has been deployed across large-scale recommendation systems spanning ads and organic surfaces, retrieval and ranking stages, and more than two orders of magnitude in inference complexity, delivering significant online gains at reduced infrastructure cost.

Dual-hand action segmentation commonly fuses left- and right-hand representations at identical temporal indices, although coordinated hand transitions may occur with nonzero and time-varying delays. We introduce Lag-Aware Cross-Hand Alignment (LACA), a lightweight module that explicitly estimates directional temporal-offset distributions between hand-specific feature streams. LACA retrieves cross-hand information from the estimated offsets and incorporates a learned null state to suppress transfer when no compatible cross-hand transition is supported. Alignment is supervised using compatibility-aware targets derived automatically from frame-level training annotations, without requiring additional labels. Analysis of the HA-ViD and ATTACH training annotations reveals robust nonzero cross-hand matches for 44.7% and 48.9% of transition anchors, respectively, compared with 18.6% and 21.3% under temporally shifted controls. When integrated into Polyphony, LACA improves the two-hand mean F1@50 from 40.4 to 42.5 and boundary F1 from 56.5 to 59.6 on HA-ViD, and from 19.9 to 21.8 and 44.7 to 47.9, respectively, on ATTACH, relative to our reproduced Polyphony baseline. These gains require only approximately 0.0086 million additional trainable parameters. We further introduce LACA-C, a future-free variant that restricts alignment and the complete inference pipeline to current and past observations. On ATTACH, LACA-C achieves 83.6% transition-cue recall, a seed-averaged median availability delay of 233~ms, 0.72 false cues per minute, and segmentation-stage throughput of 224.9 current-position predictions per second. These results demonstrate that explicit cross-hand temporal alignment improves both action segmentation and boundary localization while supporting timely future-free perception.

Common chunking strategies in Retrieval-Augmented Generation (RAG) systems often create redundant chunks. These redundant chunks make the vector database bigger and slow down retrieval. A common fix is cosine-similarity thresholding. This method reduces each chunk to a single vector, then compares vectors using a similarity score. But a single vector can lose the fine-grained, token-level detail needed to tell a true duplicate apart from a chunk that just shares the same topic. We propose Cross-Attention Calibrated Deduplication (CACD). CACD checks each new chunk against an in-memory pool of chunks already kept, using a cross-encoder instead of a single pooled vector. This keeps token-level detail all the way to the final comparison. CACD combines three parts: the cross-encoder comparison itself, a New Information Score (NIS) that measures how much of a chunk is not explained by a candidate already kept, and a majority vote across several candidates rather than a single best match. NIS is calculated from the attention entropy of the cross-encoder. We tested CACD against five existing filtering methods, nine chunking strategies, and 18 configurations, all on the full SQuAD 1.1 validation set. In our experiments, CACD removes 9.75% of chunks on average. This drop rate is close to other semantic-level methods, and much higher than exact-match filters, which barely remove anything. In these experiments, CACD also processes each configuration in 51.0 seconds on average, about 27% faster than the strongest baseline, NERExact (69.6s), and about 7x faster than cosine-similarity filtering (356.7s). These results come from a single dataset, so we present them as an early comparison, not a general claim. Code for the baseline evaluation and for CACD is available at https://github.com/lehuyphuong/rag_bench and https://github.com/lehuyphuong/cacd_dedup.

Recent 3D foundation models can generate high-quality assets from a single image, but degrade markedly on unconstrained multi-image inputs, often producing distorted geometry, over-smoothed textures, and chaotic colors. We argue that this failure stems not from limited model capacity, but from a mismatch between single-image cross-attention and the multi-image setting: existing models lack a principled way to decide which image each 3D voxel should trust at each denoising step. Revisiting recent single-image 3D foundation models, we show that explicitly routing each voxel to its most informative image is sufficient to unlock strong performance on inconsistent multi-image inputs. Based on this observation, we propose UMI3D, a training-free and plug-and-play framework that restructures cross-attention for unconstrained multi-image 3D generation. Its core, Simultaneous Focus Cross-Attention (SFC-Attn), activates all conditioning images at each denoising step while allowing each voxel to focus on the single image that best explains it. To enable this routing, we derive the Voxel Reference Score (VRS), a model-intrinsic metric for voxel--image affinity that requires no external matching, segmentation, or correspondence models. Extensive experiments show that UMI3D unlocks the multi-image potential of single-image 3D generation frameworks across diverse tasks. Project Page: UMI3D-Project.github.io.

Modeling multi-entity temporal data requires capturing dependencies across entities, time, and their interactions. Transformer-based approaches perform well but often rely on deep stacks of layers to learn these heterogeneous dependencies implicitly, increasing computational cost. We revisit this problem from a structural perspective and decompose multi-entity temporal dynamics into three interaction types: spatial interactions among entities, temporal interactions across time, and cross interactions coupling the two domains. We propose a structured spatio-temporal transformer block that explicitly models all three within a single stage. It uses parallel spatial and temporal self-attention, followed by bidirectional cross-attention, and combines the outputs through learnable gated fusion. By directly encoding these complementary views, the model reduces the need for deep stacking. We evaluate the approach on video-based group activity recognition, skeleton-based human interaction analysis, and wearable sensor-based activity recognition. Despite its simplicity, the single structured Transformer block matches or outperforms deeper architectures with only 1.76M parameters. The results suggest that depth in prior models partly compensates for implicit and entangled interaction modeling, whereas explicit factorization offers a more efficient and transparent alternative. More broadly, this work supports a structure-first design principle: expressive multi-entity temporal reasoning can emerge by exposing interaction structure rather than relying on depth.

The Segment Anything Model 2 (SAM2) has advanced temporal promptable segmentation, yet its deployment remains hindered by heavy memory cross-attention overhead and redundant full-frame visual feature extraction. While recent methods explore efficiency via heuristic memory pruning and window-based sparse routing, they typically suffer from catastrophic performance degradation in complex segmentation scenarios replete with occlusions and distractors. To resolve these limitations, we propose \textbf{Lean-SAM2}, a holistic lightweight framework designed to address the above vulnerabilities while systematically eliminating computational redundancies. Specifically, Lean-SAM2 integrates three collaborative mechanisms: (1) Target-Anchored Memory Pruning (TAMP) safeguards target tokens against deceptive attention by modulating raw attention significance with semantic consistency against prompt-derived foreground anchors; (2) Temporal Condensation with Insurance Memory (TCIM) condenses historical context via a visibility-gated fusion while conditionally archiving high-confidence entries in a parallel insurance bank; and (3) Target-Anchored Risk-Aware Routing (TARR) selectively activates the heavy image encoder for target-related windows based on anchor similarity, utilizing a risk-aware fallback policy to trigger full-frame refreshes during volatile transitions. Extensive evaluations across multiple challenging benchmarks demonstrate that Lean-SAM2 establishes a superior balance between accuracy and efficiency. For example, on the LVOSv2 validation dataset, Lean-SAM2 achieves overall inference speedups of $1.412\times$ and $1.417\times$ on the SAM2.1-Large and SAM2.1-Base+, respectively, significantly outperforming Efficient-SAM2 while boosting the corresponding $\mathcal{J}\&\mathcal{F}$ scores by $5.0\%$ and $3.6\%$. Code is available at https://github.com/DeawhaleQwQ/Lean-SAM2.

Diffusion Transformers achieve strong image generation performance, but most operate in compressed latent spaces. Pixel-space diffusion avoids this information loss, yet existing approaches map each raw image patch to a single token, forcing one representation to handle both global communication and fine-grained details. We address this issue by proposing a new architecture, \textbf{DuSPiT}, a \textbf{Du}al-branch \textbf{S}ub\textbf{P}atch \textbf{Pi}xel \textbf{T}ransformer. This model separates global structural reasoning from local appearance modeling. DuSPiT uses a compact base branch for efficient global reasoning and a parallel, high-capacity pixel branch, organized into subpatch groups, to preserve detailed appearance, with the two branches interacting through cross-attention. Our results show that DuSPiT generates images with richer details and stronger fine-grained structures, while also achieving a better quality--efficiency trade-off than prior pixel-space diffusion transformers.

Deepfake generation has raised growing concerns regarding digital media authenticity, misinformation, identity fraud, and public trust. Recent studies show that combining spatial and frequency features leads to stronger detection results than using independently. This paper presents MSCA-FFT, a Fast Fourier Transform (FFT)-based multi-scale cross-attention framework for image-level deepfake detection. The model combines a partially fine-tuned Xception spatial branch with an FFT-based frequency branch. The frequency branch processes the log-scaled FFT magnitude spectrum through shallow convolutional layers, avoiding inverse frequency-to-image reconstruction used in DCT-based pipelines. The spatial and frequency representations are refined by transformer encoders, fused through cross-attention, and passed to an MLP classifier for real/fake prediction. Experimental results show that MSCA-FFT achieves consistently higher performance than the DCT-based state-of-the-art spatial-frequency fusion method and the compared baseline models. The ablation study further indicates that the FFT-based frequency branch provides complementary spectral cues when fused with spatial features. In addition, FFT-based frequency analysis and Grad-CAM/LIME explanations show consistent evidence around manipulation-sensitive facial regions, including the eyes, mouth, nose, and facial boundaries.

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.

Machine unlearning is widely used to remove hazardous knowledge from large language models. Modern Vision-Language Models (VLMs), however, process both text and visual inputs, raising a fundamental security question: does unlearning in one modality transfer to the other? We present the first systematic, bidirectional study of cross-modal unlearning transfer across three VLM architectures: LLaVA-1.5 (MLP projection), InstructBLIP (Q-Former), and IDEFICS (gated cross-attention). We find that unlearning transfers across modalities, but the transfer is asymmetric and incomplete. In some cases, text unlearning strongly transfers to vision. However, this robustness is not preserved under typographic attacks that manipulate the visual presentation of text. Under such attacks, previously unlearned knowledge can be readily recovered, indicating shallow unlearning. To address the transfer gap and shallow robustness, we propose \textsc{CrossInf}, an influence-guided mitigation strategy. Motivated by the observation that different model components contribute unequally to cross-modal transfer, \textsc{CrossInf} focuses unlearning on transformer blocks that most influence cross-modal generalization. It reduces the transfer gap by more than half in architectures with strong fusion, while preserving model utility. It also improves robustness under typographic attacks, reducing the attack success rate to near zero. We further conduct human evaluation with three annotators ($κ{=}0.77$) to validate our findings. Finally, we analyze shallow unlearning using Centered Kernel Alignment (CKA), providing insights into the observed transfer behavior and robustness limitations.

We present a multimodal framework for Ambivalence/Hesitancy (A/H) recognition in video, developed for the ABAW11 challenge at ECCV 2026. The proposed approach fuses textual, acoustic, and visual modalities extracted from the BAH dataset using three pretrained encoders: F2LLM-v2-0.6B for transcripts (1024-d), WavLM-Large for audio (1024-d), and VideoMAE V2 for facial video (768-d). We first establish comprehensive unimodal baselines using classical classifiers (MLP, Random Forest, GBDT), each optimized via Optuna, and obtain a best unimodal Macro F1 of \textbf{0.6659} on the test set using text features alone -- substantially outperforming the zero-shot Video-LLaVA baseline (Macro F1: 0.2827). Building on these baselines, we propose a multimodal fusion architecture that combines bidirectional cross-attention across all three modalities with a Gated Multimodal Unit (GMU), with both architectural and optimization hyperparameters selected through a 50-trial Optuna search. This model achieves a Macro F1 of \textbf{0.7394} on the validation set, a relative improvement of 11.0% over the best unimodal baseline, confirming that explicit cross-modal interaction captures complementary cues that no single modality provides in isolation. Final predictions on the official, unlabeled private test set are generated using this model and submitted according to the challenge protocol. Code is publicly available at https://github.com/yassineouzar/IUSD_AH/

Low-Light Image Enhancement (LLIE) requires a careful balance among noise suppression, color fidelity, and efficiency. Recent HVI-based methods alleviate color entanglement by decoupling intensity and chromaticity, yet how reliably the two streams are fused again is an overlooked factor that largely determines the final quality. We observe that the confidence of cross-stream attention is strongly layer-dependent, so the fixed-quota selection of Top-K sparse attention is mismatched to it, discarding informative dependencies in some layers while retaining noisy ones in others. Motivated by this observation, we propose TCA-Net, a network built around Thresholded Cross-Attention that targets reliable intensity-chromaticity fusion in the HVI space rather than introducing yet another color representation. At its core, TCA replaces the rigid Top-K quota with a fixed confidence threshold whose retained cardinality is input- and layer-adaptive, retaining only high-confidence cross-stream interactions while suppressing unreliable ones. Around this core, two complementary designs clean up the fusion before and after it: a Phase-guided Fourier Interaction Module provides a structure-aware brightness initialization for the intensity stream prior to fusion, and a Decoupled Dual-Stream Guidance Module constructs residual intensity features to suppress chromaticity leakage during reconstruction. A Scale-Aware Consistency Regularization further improves structural robustness under scale perturbations during training. Extensive experiments on LOL-v1, LOL-v2, Sony-Total-Dark, and LSRW-Huawei demonstrate that TCA-Net delivers competitive restoration accuracy, improved color fidelity, and a compact parameter size.

Goal-conditioned reinforcement learning hinges on how the goal is encoded. Contrastive, metric, temporal-distance, and information-theoretic encoders differ in objective. They still share one trait. None of them sees the current state. Such a state-independent embedding cannot mark which part of the goal still needs action. The policy must then recover that cue by inverting both encoders. We propose DAGR. It refines the static embedding of any late-fusion encoder into a state-conditioned one through multi-scale gated cross-attention. A near-identity gated residual preserves the base representation. Difference-aware Goal Cross-Attention then biases the attention scores using a per-token state-goal discrepancy map. On OGBench, DAGR improves navigation. Our ablations trace the gain to the gated residual, not to the difference bias that names the method. On manipulation and puzzle tasks it matches or falls below the base. DAGR is a structured refinement, not a universal improvement.

Estimating the full shape of a deformable object is especially challenging when vision is unavailable: in the dark, inside an opaque bag, behind the manipulating hand, or under heavy self-occlusion. Touch is the natural sensor in these settings, but touches are sparse and local. We present a single topology-agnostic estimator that reconstructs the full mesh of a deformable object from only a few touches and no vision, using one permutation-invariant cross-attention architecture that handles a 1D rope, a 2D cloth, and a 3D volumetric soft body. The learned estimator reduces reconstruction error by roughly two-thirds relative to non-learned geometric mesh completion and a Gaussian-process surface baseline, and it outperforms a simpler global-pool set encoder, with the gap growing as more touches are observed. We then show that the estimator's deep-ensemble uncertainty can be used to learn where to touch next, which lowers error further and beats both random touching and a Gaussian-process active baseline at sparse budgets. This gain is modest on average but grows with self-occlusion and on the error tail. When vision is also available, where to touch barely matters, motivating the vision-free setting we study.

We present a frame-independent audio-text system for the 3rd Ambivalence/Hesitancy Video Recognition Challenge at the 11th Affective & Behavior Analysis in-the-Wild (ABAW) Workshop. Videos are divided into overlapping 5-s windows aligned with transcript timestamps. Each window combines prosodic audio descriptors, emotion-oriented RoBERTa embeddings, and 74 psycholinguistic features representing uncertainty, hedging, and attitudinal conflict. Temporal cross-attention fuses audio and text, while the support features condition gated Multiple Instance Learning (MIL) pooling. A five-seed ensemble achieves an average precision of 0.875 and a macro-F1 of 0.722 on the 525-video labeled public-test split. Notably, our submission ranked third overall on the official challenge leaderboard, with a macro-F1 of 0.7455. Source code is available at https://github.com/Liga-de-IA-PUCPR/abaw-11-ah-challenge/.

Real-world target speaker extraction (TSE) remains challenging because target speech, interference, and enrollment are recorded under mismatched acoustic conditions with reverberation, noise, and irregular conversational overlap. This paper describes the SonicAGI submission to the REAL-TSE Challenge (IEEE SLT 2026). We take a data-centric approach that combines fully simulated mixtures from clean speech with real meeting overlaps, and use a frozen offline enhancer to provide a denoised mirror of real targets for auxiliary supervision. For the online track, we introduce SwiftNet-Lookahead, which inserts a single bounded-lookahead module before a strictly causal iterative separator and keeps the total system latency at 96 ms. For the offline track, we use a frame-level enrollment cross-attention USEF-TFGridNet with a magnitude-domain fusion stage that trades off perceptual quality and speaker fidelity. In the official evaluation, SwiftNet-Lookahead ranks second in Track1 and USEF-TFGridNet ranks fifth in Track2, both exceeding the challenge baselines. These results suggest that real-data-oriented training and track-specific modeling are effective for conversational TSE.

Accurate cervical cytology image classification is a key component of automated cervical cancer screening, where reliable recognition of normal, precancerous, and cancer-associated cellular patterns from Pap smear images can improve screening efficiency and diagnostic consistency. However, this task remains challenging because cervical cells exhibit complex morphology, subtle intra-class variations, and strong inter-class similarities. Existing convolution-based models capture local texture well but have limited ability to model long-range relationships, whereas attention-based models provide broader context but often lack explicit structural guidance. To address these limitations, we propose a geometry-aware classification framework for cervical cancer screening-oriented cytology image analysis, incorporating semantic abstraction and structural priors learned from pre-trained vision-language features. The method uses Gaussian expert modules to generate axis-wise priors from global semantic information, capturing structural regularities such as nuclear alignment and cellular spatial organization. These priors are embedded into an axial self-attention module to modulate similarity computation along horizontal and vertical directions, improving long-range dependency modeling and structure-sensitive feature interaction. Experiments on the Mendeley liquid-based cytology and SIPaKMeD datasets show that the proposed method achieves 99.48% accuracy on the former and 96.08% on the latter, with balanced gains in recall, precision, and overall classification performance. Visual analysis further shows that the learned priors highlight diagnostically relevant cellular regions, demonstrating the potential of the proposed framework as a screening-oriented decision-support tool for cervical cytology.

Diffusion unlearning is essential for mitigating the generation of harmful or copyrighted content in text-to-image models. Current diffusion unlearning techniques determine the model update direction by either using alternatives of the target concept as an anchor or using empty prompts. The anchor-based method relies on manually and semantically-chosen anchors that risk biased unlearning, while the anchor-free method inherently suffers from unrobust unlearning due to unconstrained latent updates. In this work, we theoretically formalize such unstable diffusion unlearning issues under the manifold hypothesis and prove that lacking a manifold-proximal anchor inevitably induces significant normal-space drift that degrades unlearning performance. To achieve stable unlearning, we propose \mysysn, a two-stage framework that automatically synthesizes manifold-proximal anchors. However, direct geometric manifold optimization is computationally intractable. To address this challenge, \mysys introduces a novel cross-attention consistency loss which serves as a highly efficient surrogate of manifold proximity. Experimental results demonstrate that \mysys effectively achieves robust and unbiased unlearning across various state-of-the-art baselines, significantly improving targeted concept removal (by up to 31.04% in CLIP score) and non-target utility (by up to 4.18% in CLIP score). Moreover, \mysys can also be easily integrated into existing diffusion unlearning methods to enhance their unlearning performance (by 6.30% for concept removal and 6.65% for utility on average).

Deformable shape representations have proven to be robust complements to texture features in cardiac image classification, offering geometric priors that are invariant to imaging artifacts and intensity variations. However, existing deep networks perform simple concatenation to combine these distinct feature representations, which neither fully exploits their complementary nature nor learns cross-modal feature dependencies. Furthermore, this results in uniform attention across all timepoints; hence ignoring the varying diagnostic importance across the cardiac phases. In this paper, we propose a novel cardiac video classification model that, for the first time, learns temporal features in an integrated space of deformable shape and image texture representations. In particular, we design a bi-directional cross-attention in the latent space to fuse latent deformable shape and image features, allowing each modality to adaptively weight the other based on spatio-temporal correspondence. In contrast to current methods that apply uniform weighting across all the cardiac phases, our approach learns to dynamically adjust the contributions of shape and texture representations, derived from images, over time. We demonstrate state-of-the-art classification performance on a cine cardiac magnetic resonance (CMR) video dataset, achieving improved interpretability from attention mechanisms that identify diagnostically critical cardiac phases and modality contributions.

In longitudinal Alzheimer's disease (AD) diagnosis support, clinical and imaging information is often collected at irregular visits. Integrating these multimodal observations may improve diagnostic assessment, but naive fusion can degrade performance when MRI is noisy or intermittently unavailable. We propose AT-Attn, a temporal-aware multimodal framework that combines Change-and-Time encoding, time-biased asymmetric cross-attention, and gated fusion to integrate MRI with longitudinal clinical information. We evaluate AT-Attn on an MRI-retained ADNI cohort of 1,520 patients using structural MRI, six cognitive-scale trajectories, and seven static clinical variables under patient-level five-fold cross-validation. The main asymmetric AT-Attn model achieves accuracy 0.719+/-0.024, macro F1 0.721+/-0.023, ROC-AUC 0.873+/-0.013, and PR-AUC 0.783+/-0.018, outperforming unimodal and naive multimodal fusion baselines while remaining competitive with strong tabular baselines. These results suggest that a temporal-aware and constrained fusion strategy can help structural MRI contribute clinically relevant complementary information for patient-level AD diagnosis support.

We propose Bi-PT, a pipeline for reconstructing 3D four-chamber human heart meshes from clinical sparsely sampled cardiac magnetic resonance imaging (CMR) data. This work addresses the error-prone generation of 3D cardiac shape from a sparse point cloud (SPC) extracted from 2D long-axis and short-axis views used in routine clinical CMR protocols. Bi-PT enables accurate inference of the four-chamber heart mesh from the SPC by learning robust point features via bidirectional point cross-attention between an atlas and the SPC, together with per-point semantic labels that improve correspondence estimation. We formulate the deformation field as a Neural Ordinary Differential Equation (NODE) parameterized by a per-point affine transformation and translation to deform the atlas toward the target heart shape. By learning such a NODE, we can guarantee the deformation field to be a locally affine diffeomorphic deformation. We also integrate a semantic label loss into the Chamfer distance to encourage label-consistent correspondences and add a smoothness regularization to stabilize and improve the learning of the deformation field. Extensive experiments demonstrate that Bi-PT achieves accurate and robust performance compared to baselines.

Video-to-audio (V2A) generation aims to synthesize realistic audio that is both semantically consistent with and temporally synchronized to a silent video. Despite recent progress, many methods still rely on multi-stage training, resulting in high computational costs and long runtimes, or transform visual input into text to leverage pretrained text-to-audio models, sacrificing fine-grained temporal cues. To overcome these limitations, we propose Flowley, an end-to-end, single-stage training architecture that produces soundtracks by combining visual features with textual prompts. Crucially, we introduce Progressive Soft-masked Cross-Attention, which embeds audio-visual synchronization directly within its attention mechanism, adding zero additional computational cost compared to standard attention layers. We further observe that existing V2A benchmarks lack sound-oriented descriptive captions, which can potentially degrade the quality of the synthesized audio. To remedy this, we propose SoundCap, a plug-and-play pipeline for creating detailed, sound-aware captions that guide the model. Remarkably, without integrating any pretrained audio-visual alignment modules, Flowley achieves state-of-the-art performance on VGGSound across multiple metrics. Moreover, by incorporating SoundCap, we further exceed the performance of the strongest existing close-sourced methods in terms of audio quality in the zero-shot setting.

Most spoofing countermeasures now place a light classifier on top of a self-supervised (SSL) speech encoder. A growing line of work adds handcrafted acoustic features and fuses them by cross-attention, partly because the attention weights appear to explain which cues the model relies on. Two things about such fusion remain untested: whether the handcrafted features contribute anything once a strong SSL encoder is already in place, and whether the attention map faithfully reflects what the classifier actually uses. We study both questions with MOSAIC, a single model that handles logical access (LA) and physical access (PA) attacks by projecting a 152-dimensional biophonetic vector into six query tokens attending over thirteen intermediate layers of WavLM-Large. Rather than pursuing state-of-the-art accuracy, we treat the model itself as the object of analysis and apply inference-time interventions that zero out selected components. Removing the handcrafted branch lowers the EER by 0.67 percentage points under LA, so WavLM alone is sufficient there, but raises it by 0.76 points under PA, where the branch supplies replay channel evidence the encoder does not carry. The 6x13 attention map is highly stable under resampling (bootstrap rank correlation 0.99), yet its faithfulness varies by domain: layer attention weights predict the causal importance of each layer under PA (rho = +0.62) but not under LA (rho = -0.28). The handcrafted branch and its attention-based explanation are therefore trustworthy for replay attacks and not for synthetic ones. On standard benchmarks the model is mid-range on LA and ahead of the official baselines on cross-source deepfakes (6.21% EER on ASVspoof 2021 DF). The contribution is not a new architecture but a checking procedure: verify handcrafted cues and attention maps by intervention before trusting either.

Significant advancement of immersive technologies such as Virtual and Augmented Reality (VR/AR) and their integration into diverse aspects of modern life need authentication interfaces that are secure, intuitive, and compatible with embodied interaction. Traditional methods such as passwords, PINs, and device-based logins, break immersion and rely on external hardware. Recent 3D-specific behavioral approaches, such as hand-gesture, eye-tracking, and electroencephalography (EEG)-based methods, offer promising alternatives but often require specialized sensors or constrain natural movement, limiting usability in dynamic environments. We present Sign in the Air to Unlock, an in-air signature interface that enables users to authenticate by signing naturally in 3D space which is a familiar, personal, and reproducible gesture. To realize this interface, we design a point-voxel Cross-Attention Network (PV-Net) that jointly models local motion dynamics and global spatial structure from 3D trajectories. The model is evaluated on two datasets: the public DeepAirSig dataset (1,800 signatures from 40 users) and ImmAirsig, a new dataset collected using Meta Quest 2 in immersive VR (880 samples from 22 users). PV-Net achieves an Equal Error Rate of 2.5% on DeepAirSig and 76% classification accuracy on ImmAirSig. These findings highlight the potential of 3D behavioral interfaces for seamless, user-centric authentication that merges security with natural interaction in immersive environments.

Recommender systems are vital in helping users navigate vast amounts of information, offering personalized suggestions and effective explanations for these recommendations. While previous efforts have attempted to provide such explanations, evaluating their effectiveness across various scenarios remains a challenge. Enhancing these explanations is essential for improving user engagement, trust, and decision-making. To facilitate effective explanations within the recommender system, we propose a Bi-level Neural Architecture Search (Bi-NAS) framework to optimize explanations. This approach simultaneously refines cross-attention mechanisms and feature interaction functions by exploring both intra-layer and inter-layer design spaces. Furthermore, we integrate Large Language Models (LLMs) to enhance explanation generation, leveraging zero-shot prompting to produce more effective and personalized justifications. By aligning user feature preferences with item quality scores, our approach ensures that explanations reflect both user intent and item attributes, improving transparency and reasoning depth. Extensive evaluations on four real-world datasets demonstrate that Bi-NAS not only boosts recommendation accuracy but also significantly improves the effectiveness of explanations for recommender systems, providing users with clear and reliable insights into the suggestions they receive.

Text-to-image diffusion models power everyday creative tasks, but they still reproduce the demographic biases in their training data. On common prompts such as a photo of a nurse,'' a photo of a CEO'', they skew their outputs toward one gender, driven by the statistics of training data rather than anything in the text. Existing debiasing methods show promise in narrow settings but require retraining, batch-level control, or prompt-specific tuning, limiting their scalability. We propose \emph{EquiSteer}, a training-free method that works per sample by steering cross-attention (CA) activations at inference time. For each target attribute, EquiSteer precomputes steering vectors from contrastive prompts. Then at generation time, a prompt-aware gate leaves attribute-specific prompts untouched, while for neutral ones it clears existing attribute signals from the CA activations and injects a target attribute. Across SD-1.5, SD-2.1, SDXL, and SANA, EquiSteer reduces the average parity gap by up to $87\%$, with minimal effect on image quality and text-image alignment. Code is available at \href{https://github.com/Atmyre/EquiSteer}{https://github.com/Atmyre/EquiSteer}.%

Multimodal large language models (MLLMs) typically employ resampling-based projectors to transform dense visual features into a compact token sequence for language modeling. Most existing resamplers adopt a single, fixed aggregation scope via global cross-attention, which can blur fine-grained local evidence and limit the ability to capture both local details and global context within a fixed token budget. In this work, we propose MS-Resampler, a multi-scope visual resampling framework for MLLMs. MS-Resampler instantiates multiple scope-specific resamplers by injecting explicit spatial scope priors into the resampling attention, enabling each branch to aggregate visual information at a particular granularity from local to global. The outputs of these scope-specific resamplers are then adaptively fused to produce the final visual representations for language modeling. Extensive experiments on ten public multimodal benchmarks show that MS-Resampler consistently improves visual understanding and multimodal reasoning over conventional single-scope resamplers, while introducing only minimal computational overhead.

The proliferation of multimedia content on social platforms has fueled multimodal misinformation, where images are used to reinforce false claims. Consequently, Multimodal Fact-Checking (MFC) has emerged as an increasingly important research area. However, current progress is hindered by a reliance on synthetic training data and curated benchmarks that fail to capture the complexity of in-the-wild data. Furthermore, existing detection models rely on restricted intra-modality consistency or unconstrained all-to-all fusion, failing to capture nuanced relations between posts and external evidence. To address these limitations, we introduce X-POSE, a benchmark of real-world, community-annotated multimodal posts from X (formerly Twitter), augmented with full-length news articles retrieved via VLM-optimized search. Additionally, we propose TRENT, a novel MFC model that performs evidence triangulation using three parallel cross-attention streams alongside a relational fusion mechanism that explicitly models entailment and contradiction. Extensive evaluations demonstrate that TRENT consistently outperforms state-of-the-art specialized models and commercial VLMs. The code, prompt templates, and dataset are available at https://github.com/stevejpapad/evidence-triangulation

In this paper, we present the solution developed by our team, XInsight Lab, which achieved first place in Track 3 of the 4th EI-MIGA-IJCAI Challenge with a test accuracy of 0.76923. To address the challenge of weak and sparse implicit emotion evidence in long videos, this paper extends the winning solution from the previous competition and proposes a compact multi-modal temporal modeling framework. The framework integrates and evaluates the effects of multi-source features, including 2D/3D skeletons, facial expression Blendshapes, DINOv2/v3 vision foundation models, X-CLIP video features, and Gemini semantic priors. Architecturally, we propose a cross-attention mechanism that utilizes static pose features, denoted as Base, as the Query and dynamic micro-motion differential features, denoted as Offset, as the Key and Value. By capturing local relative velocities, this mechanism eliminates static biases related to individual body shape and identity. Concurrently, an adaptive pooling method based on Multiple Instance Learning is employed to extract instantaneous emotions while suppressing background noise in long sequences. Finally, the paper reveals the representation collapse phenomenon of general vision foundation models in micro-dynamic tasks, and analyzes the underlying mechanisms where networks fall into public-leaderboard-driven pseudo-generalization due to shortcut learning and rote memorization.

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

Vision-Language-Action (VLA) models have become an important paradigm of embodied AI. However, existing VLA models typically assume well-lit and stable indoor settings, while real-world embodied manipulation may involve degraded RGB observations caused by illumination shifts, posing critical challenges for robust robotic manipulation. To address this gap, we propose \textbf{Event-VLA}, an event-enhanced VLA framework for generalizable manipulation across varying illumination conditions. We formulate VLA-based manipulation under degraded visibility as a practical robustness problem for RGB-centric policies, and introduce event streams as an illumination-robust, motion-sensitive complementary observation to improve robustness across visibility levels. Specifically, unlike conventional multimodal fusion that directly merges event features into the global semantic token space, Event-VLA injects event information through an action-query routing pathway. It uses learnable action queries to extract task-relevant semantics from the VLA reasoning process, and selectively aggregates event tokens via gated cross-attention to construct event-aware action representations. This design preserves the pretrained RGB-language semantic priors while effectively leveraging event information for robust action prediction. Experiments in simulation and real-world deployment show that Event-VLA maintains strong manipulation performance under normal lighting and improves success rates under low-light degradation and near-dark real-world settings.

Intracranial aneurysms are often asymptomatic until rupture, which carries high mortality. Rupture risk assessment and treatment planning depend on both aneurysm morphology and anatomical location, yet existing automated methods remain limited to binary detection without fine-grained anatomical classification or multi-class segmentation. We present a multi-task framework that simultaneously performs multi-label classification, multi-class aneurysm segmentation, and multi-class vessel segmentation across 13 anatomical locations and four imaging modalities (CTA, MRA, T2, T1-post). Our two-stage approach combines a fast 2D tri-axial Region of Interest (ROI) extraction method with a 3D multi-task nnU-Net backbone. A dual-decoder design mitigates the extreme volume imbalance between aneurysm and vessel classes, while cross-attention pooling and modality-specific auxiliary heads improve feature learning across heterogeneous inputs. Our two-fold ensemble achieved 2nd place in the RSNA 2025 Intracranial Aneurysm Detection challenge. Code, model weights, and a 3D Slicer plugin are publicly available.

Long-term time series forecasting (LTSF) plays a crucial role in fields such as energy management, finance, and traffic prediction. Transformer-based models have adopted patch-based strategies to capture long-range dependencies, but accurately modeling shape similarities across patches and variables remains challenging due to scale differences. To address this, we introduce patch-mean decoupling (PMD), which separates the trend and residual shape information by subtracting the mean of each patch, preserving the original structure and ensuring that the attention mechanism captures true shape similarities. Futhermore, to more effectively model long-range dependencies and capture cross-variable relationships, we propose Trend Restoration Attention (TRA) and Proximal Variable Attention (PVA). The former module reintegrates the decoupled trend from PMD while calculating attention output. And the latter focuses cross-variable attention on the most relevant, recent time segments to avoid overfitting on outdated correlations. Combining these components, we propose PMDformer, a model designed to effectively capture shape similarity in long-term forecasting scenarios. Extensive experiments indicate that PMDformer outperforms existing state-of-the-art methods in stability and accuracy across multiple LTSF benchmarks. The code is available at https://github.com/aohu1105/PMDformer.

Visual prompt tuning has emerged as a parameter-efficient fine-tuning approach for adapting large-scale Vision Transformers (ViTs) to downstream tasks. As its learnable prompts are applied in input and feature spaces, prior to jointly going through attention in transformer layers, the most commonly used scheme for fusing image and prompt tokens is concatenation or addition. In this paper, we aim to study a fundamental yet essential problem in visual prompt tuning: whether a single fusion scheme tends to yield better results, and whether that would be beneficial to develop a hybrid fusion scheme. To this end, we formulate the task as a bi-level optimization problem, and solve it leveraging differentiable architecture search. In this context, the learnable prompts and their fusion schemes are jointly optimized. To enrich the search space in the architecture search, we propose two additional fusion schemes, namely, affine transformation and cross-attention, in addition to concatenation and addition. Extensive experiments on 34 datasets spanning VTAB-1k, FGVC, and HTA show consistent gains over prompt-tuning baselines. With a frozen ViT backbone, our method delivers a favorable accuracy--latency--parameter trade-off compared with VPT-Deep and recent variants. Our findings reveal that how prompts fuse with image tokens plays a significant role in visual prompt tuning, and a hybrid fusion fashion can more effectively leverage layer semantics of ViTs, contributing a novel perspective for visual prompt-tuning research.

Aligner-Encoders are recently proposed seq2seq end-to-end ASR models that replace decoder attention by predicting the uth token directly from the u-th encoder position, so the encoder must learn the alignment internally without cross-attention or a transducer lattice. In practice, this alignment often forms abruptly in the upper layers, making training sensitive and brittle on long utterances. We propose InterAligner, which adds an intermediate Aligner objective so alignment can form progressively across depth, together with an intermediate CTC loss (InterCTC) to stabilize optimization. On LibriSpeech with a 17-layer Conformer, a final-only Aligner reaches 5.0/7.8 WER (test-clean/other). InterCTC improves to 3.4/6.0, and InterAligner further reduces WER to 3.1/5.6 with the largest gains on long utterances.

Molecular surfaces encode the geometric and physicochemical patterns that determine antibody-antigen recognition, central to epitope prediction. However, existing methods rely on sequences or backbone structures and struggle to capture discontinuous, surface-driven epitopes. This study presents SurfBind, a surface-centric learning framework for epitope prediction that operates directly on molecular surface representations. SurfBind integrates geometric and physicochemical cues through a Transformer-based architecture with patch-level surface modeling, binder-aware cross-attention, and a hierarchical coarse-to-fine prediction paradigm. Experiments on challenging epitope identification benchmarks, including SAbDab and DB5.5, demonstrate that SurfBind achieves state-of-the-art performance and strong generalization across unseen antibodies and conformational states, highlighting the value of interaction-aware surface modeling for understanding the crucial mechanisms of protein-protein interactions.

As retrieval systems scale, effective and efficient reranking becomes increasingly important. However, most existing encoder- and decoder-based rerankers jointly process every query--passage pair, tightly coupling their online computation and limiting deployment efficiency and flexibility. We present KaLM-Reranker-V1, a fast but not late-interaction FBNL reranker that decouples query and passage computation while retaining expressive relevance modeling. Built on an encoder--decoder architecture, KaLM-Reranker-V1 pre-encodes passages using Matryoshka embedding pooling, while its decoder models system and user instructions together with query intent; cross-attention then captures fine-grained relevance between the resulting query context and passage representations. Together, these designs offer four key advantages: (i) efficiency from offline passage encoding, (ii) expressiveness from cross-attention, (iii) compactness from Matryoshka embedding pooling, and (iv) test-time compute through an adjustable compute budget. We instantiate KaLM-Reranker-V1 in three sizes, Nano, Small, and Large, with 0.27B, 1B, and 4B activated parameters, respectively. Extensive experiments on BEIR, MIRACL, and LMEB demonstrate strong reranking performance with superior efficiency. On BEIR and MIRACL, our models achieve competitive performance in multi-domain and multilingual reranking, on par with strong industrial rerankers such as the Qwen3/BGE-Reranker series. On LMEB-Dialogue, a compact embedding model paired with our Nano reranker, which has only 0.27B activated parameters, remains competitive with 7--12B embedding models. Data and models are available at https://huggingface.co/collections/KaLM-Embedding/lychee-kalm-reranker-and-jev.

We present OmniV2X, a generative foundation model for vehicle-to-everything (V2X) cooperative driving. The model directly interprets independent context sequences comprising multi-modal and multi-agent observations. The new design mitigates the computational cost of dense 3D perception, the vulnerability to data scarcity in cooperative scenarios, and the poor compliance with standardized messaging in existing methods that fuse multi-modal inputs into a shared representation. For training, we present an end-to-end supervised pipeline using a downstream trajectory generation loss, in which a high-capacity generative sequence planner implicitly learns to steer the model and leverage multi-modal inputs via cross-attention injection. As a foundation model, we demonstrate that OmniV2X pre-trained on large-scale single-agent planning datasets can efficiently adapt to cooperative environments by integrating the conditioning context with lightweight, standard-compliant V2X tokens. Evaluated on the DAIR-V2X-Seq dataset, OmniV2X outperforms existing end-to-end cooperative driving baselines, achieving state-of-the-art performance with less than 10% of the fine-tune V2X dataset and less than 1% of the communication bandwidth. We conduct comprehensive evaluations to demonstrate its computational efficiency and robustness under real-world constraints.