The emergency response capability and resilience of urban rail transit (URT) systems are critical to smart cities. Beyond widely studied severe disturbances, low-intensity but high-frequency daily disturbances likewise continuously challenge operational stability and can accumulate into systemic risks. However, the mechanisms and effects of such “quasi-emergency” disturbances have not been systematically studied yet. To extend research on URT resilience from severe and low-frequency disruptions to daily disturbances, this study proposes a micro–macro resilience assessment framework grounded in the passenger-level disturbance response chains. The framework integrates a fine-grained passenger itinerary inference approach incorporating daily disturbance effects, with a multi-source feature driven daily disturbance identification model. This integration overcomes the critical challenge of lacking daily disturbance data, enabling multi-dimensional resilience quantification from passenger-level micro-behaviors to system-wide macro-performance. Empirical analysis of the Chengdu Metro shows a generally favorable baseline resilience performance, with 18/31 of study days classified as good or very good. Resilience patterns are closely linked to passenger flow regularities and distributions of daily disturbances, revealing spatio-temporal heterogeneity and functional dependency. Spatially, it exhibits structural and functional weaknesses. Temporally, holidays with surging passenger demand represent the resilience troughs, with the lowest value of 0.8540. In the disturbance process dimension, delayed responses following disturbance onset and inadequate recovery during the terminal phase are identified as the primary deficiencies. Based on the numerical results, relevant management strategies and optimization measures are proposed, offering actionable insights for enhancing the disturbance-handling capacity and resilience of URT systems.

GNNs are often trained as low confident models, making it difficult for them to produce reliable estimations. We observed that graph-level GNNs also suffer from this issue, which hinders the application of graph-level GNNs in high-stake fields. Graph readout function is a common component in graph-level tasks, reducing node representations into a graph representation. This process inevitably leads to the vanishment of explicit structure. This loss of structural confidence information after readout layers prevents the use of existing calibration methods in graph-level tasks. To address this gap, we propose ConfiBR, a post-hoc, accuracy-preserving, and graph topology-aware calibration method for general graph-level GNNs. Specifically, for the logits produced by the original classification GNNs, we use a Confidence Decomposition module to calculate node-level confidence scores using a backward-readout process. These scores are then fed into a Confidence Aggregator module for training, where a graph-wise calibration factor is learned to make the prediction of the model credible, while preserving the accuracy of the original classification GNN. Experimental results on classification and regression benchmark datasets demonstrate that ConfiBR achieves state-of-the-art calibration performance. Ablation studies further validate the effectiveness and robustness of our proposed method. Code and data are available at https://github.com/33646341/ConfiBR.

Ultrasound image segmentation is essential for delineating anatomical structures and lesions, providing the foundation for accurate diagnosis. While the Segment Anything Model (SAM) has demonstrated remarkable success on natural images, its performance on ultrasound data is often hindered by poor boundary delineation. To address this limitation, we propose EP-SAM, an edge-aware and prompt-enhanced adaptation of SAM. Specifically, we leverage multi-block feature extraction from the image encoder to enrich coarse-to-fine semantic representations, while edge-aware supervision of the image encoder improves robustness to contour ambiguity and speckle noise. By integrating these complementary cues, EP-SAM generates high-quality prompts that effectively guide the model toward target regions of interest. Experimental results on multiple benchmarks demonstrate that EP-SAM consistently outperforms existing SAM-based methods.

Given the vastness of chemical space and the ongoing emergence of previously uncharacterized proteins, zero-shot compound-protein interaction (CPI) prediction better reflects the practical challenges and requirements of real-world drug development. Although existing methods perform adequately during certain CPI tasks, they still face the following challenges: (1) Representation learning from local or complete protein sequences often overlooks the complex interdependencies between subsequences, which are essential for predicting spatial structures and binding properties. (2) Dependence on large-scale or scarce multimodal protein datasets demands significant training data and computational resources, limiting scalability and efficiency. To address these challenges, we propose a novel approach that pretrains protein representations for CPI prediction tasks using subsequence reordering, explicitly capturing the dependencies between protein subsequences. Furthermore, we apply length-variable protein augmentation to ensure excellent pretraining performance on small training datasets. To evaluate the model’s effectiveness and zero-shot learning ability, we combine it with various baseline methods. The results demonstrate that our approach can improve the baseline model’s performance on the CPI task, especially in the challenging zero-shot scenario. Compared to existing pre-training models, our model demonstrates superior performance, particularly in data-scarce scenarios where training samples are limited. Our implementation is available at https://github.com/Hoch-Zhang/DrugDiscovery-DTI/.

Transformer-based methods have been popular for a variety of visual perception tasks due to their better global modeling via attention. However, a plain transformer-based architecture is known for lacking inductive biases, which will impede the performance in multi-task learning (MTL) of dense prediction due to the incapability of capturing task-relevant prior information. To this end, we propose the task prior attention network (TPANet), which introduces task-relevant prior information into the whole architecture. Our TPANet consists of three tailored modules: task prior extractor, adaptive task mixing, and cross attention modules. First, the proposed task prior extractor is applied for introducing task-relevant prior information with inductive biases via convolution for each task, adapting them to the downstream module simultaneously. Second, for task interaction efficiency, our method relies on the adaptive task mixing equipped with spatial and channel mixing to capture the task interaction. Third, the proposed cross attention module is leveraged to query task-specific feature maps with task-relevant prior information via query-based attention. Our method allows compatibility with different backbones. TPANet (with Swin-L) performance surpasses the previous state-of-the-art by a large margin of +4.6 mIoU on NYUD-v2 and +0.8 mIoU on PASCAL-Context dataset, demonstrating the potential of our method as a robust MTL model.

Remote sensing change detection identifies semantic inconsistency between spatially aligned bi-temporal observations. Most existing methods construct explicit discrepancy features, such as subtraction, addition, concatenation, or learned fusion, before decoding the change map. We revisit this design and propose SEED, a Siamese Encoder--Exchange--Decoder framework that uses parameter-free feature exchange as the only cross-temporal interaction mechanism. With shared encoder and decoder weights, SEED learns change cues directly from exchanged bi-temporal representations. We further formulate feature exchange as a permutation operator. Under pixel consistency and a fixed exchange mask, the transformation is orthogonal, invertible, and preserves paired information, whereas common arithmetic fusion operators that collapse two streams are generally non-invertible. Experiments on five public benchmarks, including SYSU-CD, LEVIR-CD, PX-CLCD, WaterCD, and CDD, and three representative backbones, including Swin Transformer V2, EfficientNet-B4, and ResNet50, show that SEED achieves competitive or superior performance with a compact and interpretable design. Studies on randomized exchange, single-decoder inference, segmentation-to-change-detection conversion, and systematic misregistration further support the flexibility and robustness of the proposed paradigm. Code and full training/evaluation protocols will be released at https://github.com/dyzy41/open-rscd.

Jailbreak attacks to Large audio-language models (LALMs) are studied recently, but they exclusively focused on the attack scenario where the adversary can fully manipulate user prompts (named strong adversary) and limited in effectiveness, applicability, and practicability. In this work, we first conduct an extensive evaluation showing that advanced text jailbreak attacks cannot be easily ported to end-to-end LALMs via text-to-speech (TTS) techniques. We then propose AUDIOJAILBREAK, a novel audio jailbreak attack, featuring (1) asynchrony: the jailbreak audios do not need to align with user prompts in the time axis by crafting suffixal jailbreak audios; (2) universality: a single jailbreak perturbation is effective for different prompts by incorporating multiple prompts into the perturbation generation; (3) stealthiness: the malicious intent of jailbreak audios is concealed by proposing various intent concealment strategies; and (4) over-the-air robustness: the jailbreak audios remain effective when being played over the air by incorporating reverberation into the perturbation generation. In contrast, all prior audio jailbreak attacks cannot offer asynchrony, universality, stealthiness, and/or over-the-air robustness. Moreover, AUDIOJAILBREAK is also applicable to a more practical and broader attack scenario where the adversary cannot fully manipulate user prompts (named weak adversary). Extensive experiments with thus far the most LALMs demonstrate the high effectiveness of AUDIOJAILBREAK, in particular, it can jailbreak openAI's GPT-4o-Audio and bypass Meta's Llama-Guard-3 safeguard, in the weak adversary scenario. We highlight that our work peeks into the security implications of audio jailbreak attacks against LALMs, and realistically fosters improving their robustness, especially for the newly proposed weak adversary.

A $k$-core is a maximal subgraph in which every vertex has degree at least $k$. As a basic task in complex network analysis, $k$-core decomposition reveals the nested sequence of all $k$-cores of an input graph for $k=1,2, \ldots, k_{\text {max }}$. As real-world graphs continue to grow in size and complexity, high-performance algorithms have become increasingly important. Recent works have leveraged single GPU to accelerate this task via a global peeling process, achieving excellent performance. However, the level-by-level peeling process suffers from limited scalability due to the long k-core sequence in real-world graphs, not only underutilizing the growing GPU compute power, but also struggling to process ever-growing graphs in multi-GPU settings due to frequent synchronization. In this paper, we exploit the inherent locality of $k$-core to propose a scalable GPU-based $k$-core decomposition framework that scales nearly linearly with single-GPU compute power and efficiently supports multi-GPU processing. Technically, we design a redundancy-free, GPU-friendly locality-aware $k$-core algorithm and then propose a local subgraph convergence mechanism to further reduce multi-GPU synchronization overhead. Our evaluation on an NVIDIA RTX PRO 6000 Blackwell GPU shows that HistCore outperforms recent single-GPU methods by 3.7× on average. On up to four RTX 3090 Ti GPUs, HistCore achieves an average speedup of $\text{1 6. 3} \times$ over NVIDIA's cuGraph and successfully processes all 8 billion-edge graphs.

Intelligent fetal ultrasound (US) interpretation is crucial for prenatal diagnosis, but high annotation costs and operator-induced variance make unsupervised pre-training a highly promising paradigm. However, existing pre-training methods largely ignore US-specific characteristics – severe data redundancy, fan-shaped locality, and polar coordinate beamforming – limiting their effectiveness in downstream tasks. To address this, we propose PolarMAE, a novel and efficient pre-training framework tailored for US images. Specifically, to mitigate continuous scanning redundancy, we introduce a Progressive Visual-Semantic Screening (PVSS) that adaptively extracts high-value samples, significantly boosting pre-training efficiency. Furthermore, we design an Acoustic-Bounded Region Constraint (ABRC) to accommodate US locality, forcing the model to focus strictly on valid acoustic regions rather than invalid dark backgrounds. Finally, leveraging the beamforming prior and local details, we propose a Polar-Texture Collaborative Masking (PTCM), enabling the model to capture underlying radial imaging patterns and critical tissue structures. Extensive experiments across diverse datasets and downstream interpretation tasks demonstrate that our method achieves state-of-the-art performance with strong pre-training scalability and efficiency.

Humanoid robot manipulation is a crucial research area for executing diverse human-level tasks, involving high-level semantic reasoning and low-level action generation. However, precise scene understanding and sample-efficient learning from human demonstrations remain critical challenges, severely hindering the applicability and generalizability of existing frameworks. This paper presents a novel RGMP-S, Recurrent Geometric-prior Multimodal Policy with Spiking features, facilitating both high-level skill reasoning and data-efficient motion synthesis. To ground high-level reasoning in physical reality, we leverage lightweight 2D geometric inductive biases to enable precise 3D scene understanding within the vision-language model. Specifically, we construct a Long-horizon Geometric Prior Skill Selector that effectively aligns the semantic instructions with spatial constraints, ultimately achieving robust generalization in unseen environments. For the data efficiency issue in robotic action generation, we introduce a Recursive Adaptive Spiking Network. We parameterize robot-object interactions via recursive spiking for spatiotemporal consistency, fully distilling long-horizon dynamic features while mitigating the overfitting issue in sparse demonstration scenarios. Extensive experiments are conducted across the Maniskill simulation benchmark and three heterogeneous real-world robotic systems, encompassing a custom-developed humanoid, a desktop manipulator, and a commercial robotic platform. Empirical results substantiate the superiority of our method over state-of-the-art baselines and validate the efficacy of the proposed modules in diverse generalization scenarios. To facilitate reproducibility, the source code and video demonstrations are publicly available at https://github.com/xtli12/RGMP-S.git.

Catastrophic forgetting, the degradation of knowledge about previously seen classes when learning new concepts from a shifting data stream, is a pitfall faced by neural network learning in open environments. Recent research on continual image classification usually relies on storing samples or prototypes to resist this forgetting. We find that during acquiring knowledge of the new classes, the features of old classes gradually disperse, which leads to confusion of features between classes and makes them difficult to discriminate. Coping with feature dispersion would be a key consideration in resisting catastrophic forgetting, which has been neglected in previous works. To this end, we try to address this issue from two perspectives. First, we propose a dispersing feature generation mechanism, which generates pseudo-features based on the pre-pooling prototypes of the old classes to simulate feature dispersion and remind the classifier to adjust the decision boundary. Second, we design a consistent alignment constraint to alleviate the severity of feature dispersion by maintaining consistency in the hidden states of different depths when aligning the current model with the previous model. Extensive experimental results on various benchmarks show the superiority of our proposed method.

Deep learning (DL) has shown remarkable effectiveness in hyperspectral image change detection (HSI-CD). Cross-domain few-shot learning (CD-FSL) has emerged as an effective solution to label scarcity in the target domain by exploiting labeled source domain data. However, existing methods often suffer from the interference of domain-sensitive channels and the instability of class prototypes. To address these issues, this article proposes a domain dropout-driven prototype enhancement network (DDPEN) for cross-domain few-shot hyperspectral change detection. The method is built upon the metric-based meta-learning to learn transferable knowledge before adapting to new HSI-CD tasks. Specifically, a dual-branch domain dropout-driven feature extractor (DDFE) is constructed to adaptively dropout domain-sensitive channels, thereby extracting domain-invariant features to mitigate domain shift. Subsequently, a history-aware prototype enhancement (HAPE) module is designed to accumulate historical and current prototype information, thus improving the representativeness and stability of class prototypes. Finally, the trained model is transferred to the target domain to construct a change detection inference module under few-shot conditions. Experiments on several hyperspectral datasets illustrate the effectiveness of the proposed algorithm.

Sparse Mixture of Experts (MoE) large language models (LLMs) are gradually becoming the mainstream approach for ultra-large-scale models. Existing optimization efforts for MoE models have focused primarily on coarse-grained MoE architectures. With the emergence of DeepSeek Models, fine-grained MoE models are gaining popularity, yet research on them remains limited. Therefore, we want to discuss the efficiency dynamic under different service loads. Additionally, fine-grained models allow deployers to reduce the number of routed experts, both activated counts and total counts, raising the question of how this reduction affects the trade-off between MoE efficiency and performance. Our findings indicate that while deploying MoE models presents greater challenges, it also offers significant optimization opportunities. Reducing the number of activated experts can lead to substantial efficiency improvements in certain scenarios, with only minor performance degradation. Reducing the total number of experts provides limited efficiency gains but results in severe performance degradation. Our method can increase throughput by at least 10% without any performance degradation. Overall, we conclude that MoE inference optimization remains an area with substantial potential for exploration and improvement.

Mars hyperspectral image classification (Mars HSIC) provides invaluable information for accurate mineral identification and Martian geology studies. However, this task remains challenging owing to scarce labeled samples, high spectral similarity, and environmental noise. Capturing both local details and global contexts is crucial for this task, yet convolutional neural networks (CNNs) struggle with global dependencies, while Transformers tend to neglect fine local details and incur unnecessary computational burden. To address this issue, we propose MMFormer, a macro-micro Transformer tailored for Mars HSIC. With a dual-branch architecture, our method adopts the macro-kernel perception and micro-kernel modulation strategy to synergize global contexts with local details in both spatial and spectral domains. In particular, the spatial branch adopts macro-kernel perception to capture long-range spatial dependencies, accompanied by micro-kernel modulation for local detail extraction. Meanwhile, the spectral branch utilizes an analogous strategy to perceive long-range spectral correlations and modulate fine-grained spectral variations simultaneously. To mitigate severe background interference, we introduce a center-prompt attention (CPA) module that injects a center-focused prompt to adaptively reweight attention toward the central pixel. To integrate these distinct feature representations, a cross-weighted fusion (CWF) module is constructed to dynamically fuse complementary spatial and spectral information. Extensive experiments on three Mars hyperspectral datasets demonstrate that MMFormer achieves strong performance and robustness under limited labeled samples and complex environmental conditions. The code is available at https://github.com/tingruifeng/TGRS_2026_MMFormer.

Recent advances in 3D Gaussian Splatting (3DGS) enable real-time, high-fidelity novel view synthesis (NVS) with explicit 3D representations. However, performance degradation and instability remain significant under sparse-view conditions. In this work, we identify two key failure modes under sparse-view conditions: overfitting in regions with excessive Gaussian density near the camera, and underfitting in distant areas with insufficient Gaussian coverage. To address these challenges, we propose a unified framework \modelname{}, comprising two key components: a Depth-and-Density Guided Dropout strategy that suppresses overfitting by adaptively masking redundant Gaussians based on density and depth, and a Distance-Aware Fidelity Enhancement module that improves reconstruction quality in under-fitted far-field areas through targeted supervision. Moreover, we introduce a new evaluation metric to quantify the stability of learned Gaussian distributions, providing insights into the robustness of the sparse-view 3DGS. Extensive experiments on multiple datasets demonstrate that our method significantly improves both visual quality and robustness under sparse view conditions. The source code and trained models will be made publicly available.

Ultrasound image classification is essential for computer-aided diagnosis. However, current methods often neglect clinical priors, leading to poor generalization in challenging scenarios and a lack of interpretability that limits clinical adoption. To address these issues, we aim to develop a medical-prior module that can be seamlessly integrated into existing pipelines to enhance both diagnostic performance and interpretability. In this paper, we propose an attribute-guided dual-branch framework for ultrasound classification that introduces domain-agnostic medical attribute priors, improving generalization while offering interpretable evidence. Specifically, a baseline branch follows conventional architectures and predicts image categories via a fully connected classifier. An attribute-guided branch injects domain-agnostic attributes as priors and produces human-interpretable decision cues. Finally, an adaptive decision module fuses the two branches in a data-dependent manner to yield the final prediction. Experiments across diverse ultrasound classification tasks demonstrate that our approach can be integrated into multiple backbones and state-of-the-art methods with low overhead, consistently improving accuracy and interpretability. Code is available at: https://github.com/zhaobo253-crypto/AttrGuide.

Power transmission corridor hazard segmentation (PTCHS) aims to separate transmission equipment and surrounding hazards from complex backgrounds, conveying great significance to maintaining electric power transmission safety. Recently, the Segment Anything Model (SAM) has emerged as a foundational vision model and pushed the boundaries of segmentation tasks. However, SAM struggles to deal with the target objects in complex transmission corridor scenario, especially those with fine structures. In this paper, we propose ELE-SAM, adapting SAM for the PTCHS task. Technically, we develop a Context-Aware Prompt Adapter to achieve better prompt tokens via incorporating global-local features and focusing more on key regions. Subsequently, to tackle the hazard objects with fine structures in complex background, we design a High-Fidelity Mask Decoder by leveraging multi-granularity mask features and then scaling them to a higher resolution. Moreover, to train ELE-SAM and advance this field, we construct the ELE-40 K benchmark, the first large-scale and real-world dataset for PTCHS including 44,094 image-mask pairs. Experimental results for ELE-40 K demonstrate the superior performance that ELE-SAM outperforms the baseline model with an average 16.8% in mIoU and 20.6% mBIoU performance improvement. Moreover, compared with the state-of-the-art method on HQSeg-44 K, the average 2.9% mIoU and 3.8% mBIoU absolute improvements further validate the effectiveness of our method on high-quality generic object segmentation. The source code and dataset are available at https://github.com/Hhaizee/ELE-SAM.