Zhiyong Wang

Zhenfei Yin, Jiong Wang, Jianjian Cao et al.

Large language models have emerged as a promising approach towards achieving general-purpose AI agents. The thriving open-source LLM community has greatly accelerated the development of agents that support human-machine dialogue interaction through natural language processing. However, human interaction with the world extends beyond only text as a modality, and other modalities such as vision are also crucial. Recent works on multi-modal large language models, such as GPT-4V and Bard, have demonstrated their effectiveness in handling visual modalities. However, the transparency of these works is limited and insufficient to support academic research. To the best of our knowledge, we present one of the very first open-source endeavors in the field, LAMM, encompassing a Language-Assisted Multi-Modal instruction tuning dataset, framework, and benchmark. Our aim is to establish LAMM as a growing ecosystem for training and evaluating MLLMs, with a specific focus on facilitating AI agents capable of bridging the gap between ideas and execution, thereby enabling seamless human-AI interaction. Our main contribution is three-fold: 1) We present a comprehensive dataset and benchmark, which cover a wide range of vision tasks for 2D and 3D vision. Extensive experiments validate the effectiveness of our dataset and benchmark. 2) We outline the detailed methodology of constructing multi-modal instruction tuning datasets and benchmarks for MLLMs, enabling rapid scaling and extension of MLLM research to diverse domains, tasks, and modalities. 3) We provide a primary but potential MLLM training framework optimized for modality extension. We also provide baseline models, comprehensive experimental observations, and analysis to accelerate future research. Our baseline model is trained within 24 A100 GPU hours, framework supports training with V100 and RTX3090 is available thanks to the open-source society.

Wenxi Yue, Jing Zhang, Kun Hu et al.

The Segment Anything Model (SAM) is a powerful foundation model that has revolutionised image segmentation. To apply SAM to surgical instrument segmentation, a common approach is to locate precise points or boxes of instruments and then use them as prompts for SAM in a zero-shot manner. However, we observe two problems with this naive pipeline: (1) the domain gap between natural objects and surgical instruments leads to inferior generalisation of SAM; and (2) SAM relies on precise point or box locations for accurate segmentation, requiring either extensive manual guidance or a well-performing specialist detector for prompt preparation, which leads to a complex multi-stage pipeline. To address these problems, we introduce SurgicalSAM, a novel end-to-end efficient-tuning approach for SAM to effectively integrate surgical-specific information with SAM's pre-trained knowledge for improved generalisation. Specifically, we propose a lightweight prototype-based class prompt encoder for tuning, which directly generates prompt embeddings from class prototypes and eliminates the use of explicit prompts for improved robustness and a simpler pipeline. In addition, to address the low inter-class variance among surgical instrument categories, we propose contrastive prototype learning, further enhancing the discrimination of the class prototypes for more accurate class prompting. The results of extensive experiments on both EndoVis2018 and EndoVis2017 datasets demonstrate that SurgicalSAM achieves state-of-the-art performance while only requiring a small number of tunable parameters. The source code is available at https://github.com/wenxi-yue/SurgicalSAM.

Rusheng Pan, Zhiyong Wang, Yating Wei et al.

A computational graph in a deep neural network (DNN) denotes a specific data flow diagram (DFD) composed of many tensors and operators. Existing toolkits for visualizing computational graphs are not applicable when the structure is highly complicated and large-scale (e.g., BERT [1]). To address this problem, we propose leveraging a suite of visual simplification techniques, including a cycle-removing method, a module-based edge-pruning algorithm, and an isomorphic subgraph stacking strategy. We design and implement an interactive visualization system that is suitable for computational graphs with up to 10 thousand elements. Experimental results and usage scenarios demonstrate that our tool reduces 60% elements on average and hence enhances the performance for recognizing and diagnosing DNN models. Our contributions are integrated into an open-source DNN visualization toolkit, namely, MindInsight [2].

Qiuxia Wu, Jiarui Lan, Wenxiong Kang et al.

Recognizing human actions from point cloud sequences is critical for 3D perception driven applications such as autonomous driving and human-computer interaction. However, the irregular structure and temporal inconsistency of point clouds pose unique challenges for spatio-temporal representation learning, especially in capturing both global motion context and fine-grained temporal dynamics. We propose SRENet, a spectral-aware framework designed to explicitly learn both global context and fine-grained temporal dynamics of motion from a frequency perspective for action recognition. SRENet introduces a Spectral Decomposition Block (SDeBlock) that performs wavelet-based analysis along temporal and spatial axes, disentangling features into low- and high-frequency components with frequency-specific attention. To recover residual dynamics and re-align temporal frequency structures distorted during semantic fusion, a Spectral Re-entry Block (SReBlock) performs secondary temporal decomposition. Furthermore, a spectral-aware learning strategy is devised to enhance discriminability in both frequency subspaces via contrastive loss and a curriculum schedule that gradually shifts focus from low- to high-frequency spaces in line with coarse to detailed motion patterns. Extensive experiments on MSR-Action3D, NTU-RGBD and NTU-RGBD120 demonstrate that SRENet achieves state-of-the-art performance, validating the effectiveness of frequency modeling in point cloud-based action understanding.

Zhiyong Wang, Xutong Liu, Shuai Li et al. · uw

Conversational contextual bandits elicit user preferences by occasionally querying for explicit feedback on key-terms to accelerate learning. However, there are aspects of existing approaches which limit their performance. First, information gained from key-term-level conversations and arm-level recommendations is not appropriately incorporated to speed up learning. Second, it is important to ask explorative key-terms to quickly elicit the user's potential interests in various domains to accelerate the convergence of user preference estimation, which has never been considered in existing works. To tackle these issues, we first propose ``ConLinUCB", a general framework for conversational bandits with better information incorporation, combining arm-level and key-term-level feedback to estimate user preference in one step at each time. Based on this framework, we further design two bandit algorithms with explorative key-term selection strategies, ConLinUCB-BS and ConLinUCB-MCR. We prove tighter regret upper bounds of our proposed algorithms. Particularly, ConLinUCB-BS achieves a regret bound of $O(d\sqrt{T\log T})$, better than the previous result $O(d\sqrt{T}\log T)$. Extensive experiments on synthetic and real-world data show significant advantages of our algorithms in learning accuracy (up to 54\% improvement) and computational efficiency (up to 72\% improvement), compared to the classic ConUCB algorithm, showing the potential benefit to recommender systems.

Xuelian Cheng, Yiran Zhong, Mehrtash Harandi et al. · ibm-research

The self-attention mechanism, successfully employed with the transformer structure is shown promise in many computer vision tasks including image recognition, and object detection. Despite the surge, the use of the transformer for the problem of stereo matching remains relatively unexplored. In this paper, we comprehensively investigate the use of the transformer for the problem of stereo matching, especially for laparoscopic videos, and propose a new hybrid deep stereo matching framework (HybridStereoNet) that combines the best of the CNN and the transformer in a unified design. To be specific, we investigate several ways to introduce transformers to volumetric stereo matching pipelines by analyzing the loss landscape of the designs and in-domain/cross-domain accuracy. Our analysis suggests that employing transformers for feature representation learning, while using CNNs for cost aggregation will lead to faster convergence, higher accuracy and better generalization than other options. Our extensive experiments on Sceneflow, SCARED2019 and dVPN datasets demonstrate the superior performance of our HybridStereoNet.

Peng Xia, Xingtong Yu, Ming Hu et al.

Object categories are typically organized into a multi-granularity taxonomic hierarchy. When classifying categories at different hierarchy levels, traditional uni-modal approaches focus primarily on image features, revealing limitations in complex scenarios. Recent studies integrating Vision-Language Models (VLMs) with class hierarchies have shown promise, yet they fall short of fully exploiting the hierarchical relationships. These efforts are constrained by their inability to perform effectively across varied granularity of categories. To tackle this issue, we propose a novel framework (HGCLIP) that effectively combines CLIP with a deeper exploitation of the Hierarchical class structure via Graph representation learning. We explore constructing the class hierarchy into a graph, with its nodes representing the textual or image features of each category. After passing through a graph encoder, the textual features incorporate hierarchical structure information, while the image features emphasize class-aware features derived from prototypes through the attention mechanism. Our approach demonstrates significant improvements on 11 diverse visual recognition benchmarks. Our codes are fully available at https://github.com/richard-peng-xia/HGCLIP.

Peggy Tang, Kun Hu, Lei Zhang et al.

Multimodal summarisation with multimodal output is drawing increasing attention due to the rapid growth of multimedia data. While several methods have been proposed to summarise visual-text contents, their multimodal outputs are not succinct enough at an extreme level to address the information overload issue. To the end of extreme multimodal summarisation, we introduce a new task, eXtreme Multimodal Summarisation with Multimodal Output (XMSMO) for the scenario of TL;DW - Too Long; Didn't Watch, akin to TL;DR. XMSMO aims to summarise a video-document pair into a summary with an extremely short length, which consists of one cover frame as the visual summary and one sentence as the textual summary. We propose a novel unsupervised Hierarchical Optimal Transport Network (HOT-Net) consisting of three components: hierarchical multimodal encoders, hierarchical multimodal fusion decoders, and optimal transport solvers. Our method is trained, without using reference summaries, by optimising the visual and textual coverage from the perspectives of the distance between the semantic distributions under optimal transport plans. To facilitate the study on this task, we collect a large-scale dataset XMSMO-News by harvesting 4,891 video-document pairs. The experimental results show that our method achieves promising performance in terms of ROUGE and IoU metrics.

Lintao Wang, Kun Hu, Lei Bai et al.

Synthesizing controllable motion for a character using deep learning has been a promising approach due to its potential to learn a compact model without laborious feature engineering. To produce dynamic motion from weak control signals such as desired paths, existing methods often require auxiliary information such as phases for alleviating motion ambiguity, which limits their generalisation capability. As past poses often contain useful auxiliary hints, in this paper, we propose a task-agnostic deep learning method, namely Multi-scale Control Signal-aware Transformer (MCS-T), with an attention based encoder-decoder architecture to discover the auxiliary information implicitly for synthesizing controllable motion without explicitly requiring auxiliary information such as phase. Specifically, an encoder is devised to adaptively formulate the motion patterns of a character's past poses with multi-scale skeletons, and a decoder driven by control signals to further synthesize and predict the character's state by paying context-specialised attention to the encoded past motion patterns. As a result, it helps alleviate the issues of low responsiveness and slow transition which often happen in conventional methods not using auxiliary information. Both qualitative and quantitative experimental results on an existing biped locomotion dataset, which involves diverse types of motion transitions, demonstrate the effectiveness of our method. In particular, MCS-T is able to successfully generate motions comparable to those generated by the methods using auxiliary information.

Changjie Qiu, Zhiyong Wang, Xiuhong Lin et al.

Existing point cloud modeling datasets primarily express the modeling precision by pose or trajectory precision rather than the point cloud modeling effect itself. Under this demand, we first independently construct a set of LiDAR system with an optical stage, and then we build a HPMB dataset based on the constructed LiDAR system, a High-Precision, Multi-Beam, real-world dataset. Second, we propose an modeling evaluation method based on HPMB for object-level modeling to overcome this limitation. In addition, the existing point cloud modeling methods tend to generate continuous skeletons of the global environment, hence lacking attention to the shape of complex objects. To tackle this challenge, we propose a novel learning-based joint framework, DSMNet, for high-precision 3D surface modeling from sparse point cloud frames. DSMNet comprises density-aware Point Cloud Registration (PCR) and geometry-aware Point Cloud Sampling (PCS) to effectively learn the implicit structure feature of sparse point clouds. Extensive experiments demonstrate that DSMNet outperforms the state-of-the-art methods in PCS and PCR on Multi-View Partial Point Cloud (MVP) database. Furthermore, the experiments on the open source KITTI and our proposed HPMB datasets show that DSMNet can be generalized as a post-processing of Simultaneous Localization And Mapping (SLAM), thereby improving modeling precision in environments with sparse point clouds.

Kun Hu, Shaohui Mei, Wei Wang et al.

Freezing of gait (FoG) is one of the most common symptoms of Parkinson's disease, which is a neurodegenerative disorder of the central nervous system impacting millions of people around the world. To address the pressing need to improve the quality of treatment for FoG, devising a computer-aided detection and quantification tool for FoG has been increasingly important. As a non-invasive technique for collecting motion patterns, the footstep pressure sequences obtained from pressure sensitive gait mats provide a great opportunity for evaluating FoG in the clinic and potentially in the home environment. In this study, FoG detection is formulated as a sequential modelling task and a novel deep learning architecture, namely Adversarial Spatio-temporal Network (ASTN), is proposed to learn FoG patterns across multiple levels. A novel adversarial training scheme is introduced with a multi-level subject discriminator to obtain subject-independent FoG representations, which helps to reduce the over-fitting risk due to the high inter-subject variance. As a result, robust FoG detection can be achieved for unseen subjects. The proposed scheme also sheds light on improving subject-level clinical studies from other scenarios as it can be integrated with many existing deep architectures. To the best of our knowledge, this is one of the first studies of footstep pressure-based FoG detection and the approach of utilizing ASTN is the first deep neural network architecture in pursuit of subject-independent representations. Experimental results on 393 trials collected from 21 subjects demonstrate encouraging performance of the proposed ASTN for FoG detection with an AUC 0.85.

Jialian Li, Jingyi Zhang, Zhiyong Wang et al.

Existing motion capture datasets are largely short-range and cannot yet fit the need of long-range applications. We propose LiDARHuman26M, a new human motion capture dataset captured by LiDAR at a much longer range to overcome this limitation. Our dataset also includes the ground truth human motions acquired by the IMU system and the synchronous RGB images. We further present a strong baseline method, LiDARCap, for LiDAR point cloud human motion capture. Specifically, we first utilize PointNet++ to encode features of points and then employ the inverse kinematics solver and SMPL optimizer to regress the pose through aggregating the temporally encoded features hierarchically. Quantitative and qualitative experiments show that our method outperforms the techniques based only on RGB images. Ablation experiments demonstrate that our dataset is challenging and worthy of further research. Finally, the experiments on the KITTI Dataset and the Waymo Open Dataset show that our method can be generalized to different LiDAR sensor settings.

Zhen Yu, Toan Nguyen, Yaniv Gal et al.

In practice, many medical datasets have an underlying taxonomy defined over the disease label space. However, existing classification algorithms for medical diagnoses often assume semantically independent labels. In this study, we aim to leverage class hierarchy with deep learning algorithms for more accurate and reliable skin lesion recognition. We propose a hyperbolic network to learn image embeddings and class prototypes jointly. The hyperbola provably provides a space for modeling hierarchical relations better than Euclidean geometry. Meanwhile, we restrict the distribution of hyperbolic prototypes with a distance matrix that is encoded from the class hierarchy. Accordingly, the learned prototypes preserve the semantic class relations in the embedding space and we can predict the label of an image by assigning its feature to the nearest hyperbolic class prototype. We use an in-house skin lesion dataset which consists of around 230k dermoscopic images on 65 skin diseases to verify our method. Extensive experiments provide evidence that our model can achieve higher accuracy with less severe classification errors than models without considering class relations.

Sicheng Liu, Lintao Wang, Xiaogang Zhu et al.

Extreme Multimodal Summarization with Multimodal Output (XMSMO) becomes an attractive summarization approach by integrating various types of information to create extremely concise yet informative summaries for individual modalities. Existing methods overlook the issue that multimodal data often contains more topic irrelevant information, which can mislead the model into producing inaccurate summaries especially for extremely short ones. In this paper, we propose SITransformer, a Shared Information-guided Transformer for extreme multimodal summarization. It has a shared information guided pipeline which involves a cross-modal shared information extractor and a cross-modal interaction module. The extractor formulates semantically shared salient information from different modalities by devising a novel filtering process consisting of a differentiable top-k selector and a shared-information guided gating unit. As a result, the common, salient, and relevant contents across modalities are identified. Next, a transformer with cross-modal attentions is developed for intra- and inter-modality learning with the shared information guidance to produce the extreme summary. Comprehensive experiments demonstrate that SITransformer significantly enhances the summarization quality for both video and text summaries for XMSMO. Our code will be publicly available at https://github.com/SichengLeoLiu/MMAsia24-XMSMO.

Zhiyong Wang, Jize Xie, Xutong Liu et al. · uw

The contextual linear bandit is an important online learning problem where given arm features, a learning agent selects an arm at each round to maximize the cumulative rewards in the long run. A line of works, called the clustering of bandits (CB), utilize the collaborative effect over user preferences and have shown significant improvements over classic linear bandit algorithms. However, existing CB algorithms require well-specified linear user models and can fail when this critical assumption does not hold. Whether robust CB algorithms can be designed for more practical scenarios with misspecified user models remains an open problem. In this paper, we are the first to present the important problem of clustering of bandits with misspecified user models (CBMUM), where the expected rewards in user models can be perturbed away from perfect linear models. We devise two robust CB algorithms, RCLUMB and RSCLUMB (representing the learned clustering structure with dynamic graph and sets, respectively), that can accommodate the inaccurate user preference estimations and erroneous clustering caused by model misspecifications. We prove regret upper bounds of $O(ε_*T\sqrt{md\log T} + d\sqrt{mT}\log T)$ for our algorithms under milder assumptions than previous CB works (notably, we move past a restrictive technical assumption on the distribution of the arms), which match the lower bound asymptotically in $T$ up to logarithmic factors, and also match the state-of-the-art results in several degenerate cases. The techniques in proving the regret caused by misclustering users are quite general and may be of independent interest. Experiments on both synthetic and real-world data show our outperformance over previous algorithms.

Clinton Ansun Mo, Kun Hu, Chengjiang Long et al.

Deriving sophisticated 3D motions from sparse keyframes is a particularly challenging problem, due to continuity and exceptionally skeletal precision. The action features are often derivable accurately from the full series of keyframes, and thus, leveraging the global context with transformers has been a promising data-driven embedding approach. However, existing methods are often with inputs of interpolated intermediate frame for continuity using basic interpolation methods with keyframes, which result in a trivial local minimum during training. In this paper, we propose a novel framework to formulate latent motion manifolds with keyframe-based constraints, from which the continuous nature of intermediate token representations is considered. Particularly, our proposed framework consists of two stages for identifying a latent motion subspace, i.e., a keyframe encoding stage and an intermediate token generation stage, and a subsequent motion synthesis stage to extrapolate and compose motion data from manifolds. Through our extensive experiments conducted on both the LaFAN1 and CMU Mocap datasets, our proposed method demonstrates both superior interpolation accuracy and high visual similarity to ground truth motions.

Xiangyi Wang, Pingchen Lu, Jie Mao et al.

Contextual dueling bandits form a cornerstone of preference-based decision-making, with critical applications in recommender systems and large language model alignment. However, standard algorithms rely on the idealized assumption of immediate feedback, a condition frequently violated in real-world scenarios such as prompt optimization. This setting introduces a unique theoretical challenge: unlike linear bandits, dueling bandit estimators lack closed-form solutions, rendering naive adaptations of standard weighting techniques biased. To address this, we formalize the problem of Contextual Dueling Bandits with Stochastic Delayed Feedback and propose two novel algorithms: Linear (LDB-DF) and Neural (NDB-DF) Dueling Bandits with Delayed Feedback. Central to our approach is a novel estimator that integrates an Inverse Probability Weighting (IPW) mechanism directly into the loss function, ensuring unbiased correction for delayed or missing feedback. We provide comprehensive theoretical analysis, establishing an O(d*sqrt(T)) regret bound for the linear setting and sub-linear guarantees for the neural setting. Extensive experiments on both simulated and real-world datasets demonstrate the effectiveness of our propose.

Jiarun Zhu, Yijun Hong, Xiaoquan Sun et al.

Vision-language-action (VLA) models provide a promising foundation for general-purpose robotics. However, their successful deployment in real-world scenarios requires the ability to continually acquire new skills while retaining previously learned behaviors. While pioneering research has studied the continual learning of VLA models in narrowly simulated environments, this challenge remains largely unexplored under realistic conditions. To address this limitation, we construct a real-world continual learning dataset comprising four sequential manipulation tasks, spanning rigid-object pick-and-place, contact-rich pressing, and deformable-object folding. Using this dataset, we conduct comprehensive experiments and find that VLA models suffer significant catastrophic forgetting when continually learning from heterogeneous real-world demonstrations. We then systematically evaluate experience replay and uncover key implementation factors that govern its success. In summary, this work provides the first empirical study of real-world continual VLA learning and offers practical guidance for deploying long-lived robot policies.

Kunming Su, Qiuxia Wu, Panpan Cai et al.

Masked point modeling methods have recently achieved great success in self-supervised learning for point cloud data. However, these methods are sensitive to rotations and often exhibit sharp performance drops when encountering rotational variations. In this paper, we propose a novel Rotation-Invariant Masked AutoEncoders (RI-MAE) to address two major challenges: 1) achieving rotation-invariant latent representations, and 2) facilitating self-supervised reconstruction in a rotation-invariant manner. For the first challenge, we introduce RI-Transformer, which features disentangled geometry content, rotation-invariant relative orientation and position embedding mechanisms for constructing rotation-invariant point cloud latent space. For the second challenge, a novel dual-branch student-teacher architecture is devised. It enables the self-supervised learning via the reconstruction of masked patches within the learned rotation-invariant latent space. Each branch is based on an RI-Transformer, and they are connected with an additional RI-Transformer predictor. The teacher encodes all point patches, while the student solely encodes unmasked ones. Finally, the predictor predicts the latent features of the masked patches using the output latent embeddings from the student, supervised by the outputs from the teacher. Extensive experiments demonstrate that our method is robust to rotations, achieving the state-of-the-art performance on various downstream tasks. Our code is available at https://github.com/kunmingsu07/RI-MAE.

Zhuqiang Lu, Kun Hu, Chaoyue Wang et al.

A 360-degree (omni-directional) image provides an all-encompassing spherical view of a scene. Recently, there has been an increasing interest in synthesising 360-degree images from conventional narrow field of view (NFoV) images captured by digital cameras and smartphones, for providing immersive experiences in various scenarios such as virtual reality. Yet, existing methods typically fall short in synthesizing intricate visual details or ensure the generated images align consistently with user-provided prompts. In this study, autoregressive omni-aware generative network (AOG-Net) is proposed for 360-degree image generation by out-painting an incomplete 360-degree image progressively with NFoV and text guidances joinly or individually. This autoregressive scheme not only allows for deriving finer-grained and text-consistent patterns by dynamically generating and adjusting the process but also offers users greater flexibility to edit their conditions throughout the generation process. A global-local conditioning mechanism is devised to comprehensively formulate the outpainting guidance in each autoregressive step. Text guidances, omni-visual cues, NFoV inputs and omni-geometry are encoded and further formulated with cross-attention based transformers into a global stream and a local stream into a conditioned generative backbone model. As AOG-Net is compatible to leverage large-scale models for the conditional encoder and the generative prior, it enables the generation to use extensive open-vocabulary text guidances. Comprehensive experiments on two commonly used 360-degree image datasets for both indoor and outdoor settings demonstrate the state-of-the-art performance of our proposed method. Our code will be made publicly available.

Penghui Wen, Kun Hu, Wenxi Yue et al.

Robust audio anti-spoofing has been increasingly challenging due to the recent advancements on deepfake techniques. While spectrograms have demonstrated their capability for anti-spoofing, complementary information presented in multi-order spectral patterns have not been well explored, which limits their effectiveness for varying spoofing attacks. Therefore, we propose a novel deep learning method with a spectral fusion-reconstruction strategy, namely S2pecNet, to utilise multi-order spectral patterns for robust audio anti-spoofing representations. Specifically, spectral patterns up to second-order are fused in a coarse-to-fine manner and two branches are designed for the fine-level fusion from the spectral and temporal contexts. A reconstruction from the fused representation to the input spectrograms further reduces the potential fused information loss. Our method achieved the state-of-the-art performance with an EER of 0.77% on a widely used dataset: ASVspoof2019 LA Challenge.

Zhiyong Wang, Jize Xie, Tong Yu et al.

In real-world online web systems, multiple users usually arrive sequentially into the system. For applications like click fraud and fake reviews, some users can maliciously perform corrupted (disrupted) behaviors to trick the system. Therefore, it is crucial to design efficient online learning algorithms to robustly learn from potentially corrupted user behaviors and accurately identify the corrupted users in an online manner. Existing works propose bandit algorithms robust to adversarial corruption. However, these algorithms are designed for a single user, and cannot leverage the implicit social relations among multiple users for more efficient learning. Moreover, none of them consider how to detect corrupted users online in the multiple-user scenario. In this paper, we present an important online learning problem named LOCUD to learn and utilize unknown user relations from disrupted behaviors to speed up learning, and identify the corrupted users in an online setting. To robustly learn and utilize the unknown relations among potentially corrupted users, we propose a novel bandit algorithm RCLUB-WCU. To detect the corrupted users, we devise a novel online detection algorithm OCCUD based on RCLUB-WCU's inferred user relations. We prove a regret upper bound for RCLUB-WCU, which asymptotically matches the lower bound with respect to $T$ up to logarithmic factors, and matches the state-of-the-art results in degenerate cases. We also give a theoretical guarantee for the detection accuracy of OCCUD. With extensive experiments, our methods achieve superior performance over previous bandit algorithms and high corrupted user detection accuracy.

Yudi Dai, Zhiyong Wang, Xiping Lin et al.

We introduce HiSC4D, a novel Human-centered interaction and 4D Scene Capture method, aimed at accurately and efficiently creating a dynamic digital world, containing large-scale indoor-outdoor scenes, diverse human motions, rich human-human interactions, and human-environment interactions. By utilizing body-mounted IMUs and a head-mounted LiDAR, HiSC4D can capture egocentric human motions in unconstrained space without the need for external devices and pre-built maps. This affords great flexibility and accessibility for human-centered interaction and 4D scene capturing in various environments. Taking into account that IMUs can capture human spatially unrestricted poses but are prone to drifting for long-period using, and while LiDAR is stable for global localization but rough for local positions and orientations, HiSC4D employs a joint optimization method, harmonizing all sensors and utilizing environment cues, yielding promising results for long-term capture in large scenes. To promote research of egocentric human interaction in large scenes and facilitate downstream tasks, we also present a dataset, containing 8 sequences in 4 large scenes (200 to 5,000 $m^2$), providing 36k frames of accurate 4D human motions with SMPL annotations and dynamic scenes, 31k frames of cropped human point clouds, and scene mesh of the environment. A variety of scenarios, such as the basketball gym and commercial street, alongside challenging human motions, such as daily greeting, one-on-one basketball playing, and tour guiding, demonstrate the effectiveness and the generalization ability of HiSC4D. The dataset and code will be publicated on www.lidarhumanmotion.net/hisc4d available for research purposes.

Shyam Sundhar Ramesh, Xiaotong Ji, Matthieu Zimmer et al.

RL-based post-training with GRPO is widely used to improve large language models on individual reasoning tasks. However, real-world deployment requires reliable performance across diverse tasks. A straightforward multi-task adaptation of GRPO often leads to imbalanced outcomes, with some tasks dominating optimization while others stagnate. Moreover, tasks can vary widely in how frequently prompts yield zero advantages (and thus zero gradients), which further distorts their effective contribution to the optimization signal. To address these issues, we propose a novel Multi-Task GRPO (MT-GRPO) algorithm that (i) dynamically adapts task weights to explicitly optimize worst-task performance and promote balanced progress across tasks, and (ii) introduces a ratio-preserving sampler to ensure task-wise policy gradients reflect the adapted weights. Experiments on both 3-task and 9-task settings show that MT-GRPO consistently outperforms baselines in worst-task accuracy. In particular, MT-GRPO achieves 16-28% and 6% absolute improvement on worst-task performance over standard GRPO and DAPO, respectively, while maintaining competitive average accuracy. Moreover, MT-GRPO requires 50% fewer training steps to reach 50% worst-task accuracy in the 3-task setting, demonstrating substantially improved efficiency in achieving reliable performance across tasks.

Haoyu Ji, Xueting Liu, Yu Gao et al.

Skeleton-based Temporal Action Segmentation (STAS) aims to densely parse untrimmed skeletal sequences into frame-level action categories. However, existing methods, while proficient at capturing spatio-temporal kinematics, neglect the underlying physical dynamics that govern human motion. This oversight limits inter-class discriminability between actions with similar kinematics but distinct dynamic intents, and hinders precise boundary localization where dynamic force profiles shift. To address these, we propose the Lagrangian-Dynamic Informed Network (LaDy), a framework integrating principles of Lagrangian dynamics into the segmentation process. Specifically, LaDy first computes generalized coordinates from joint positions and then estimates Lagrangian terms under physical constraints to explicitly synthesize the generalized forces. To further ensure physical coherence, our Energy Consistency Loss enforces the work-energy theorem, aligning kinetic energy change with the work done by the net force. The learned dynamics then drive a Spatio-Temporal Modulation module: Spatially, generalized forces are fused with spatial representations to provide more discriminative semantics. Temporally, salient dynamic signals are constructed for temporal gating, thereby significantly enhancing boundary awareness. Experiments on challenging datasets show that LaDy achieves state-of-the-art performance, validating the integration of physical dynamics for action segmentation. Code is available at https://github.com/HaoyuJi/LaDy.

Haoyu Ji, Bowen Chen, Zhihao Yang et al.

Skeleton-based Temporal Action Segmentation (STAS) seeks to densely segment and classify diverse actions within long, untrimmed skeletal motion sequences. However, existing STAS methodologies face challenges of limited inter-class discriminability and blurred segmentation boundaries, primarily due to insufficient distinction of spatio-temporal patterns between adjacent actions. To address these limitations, we propose Spectral Scalpel, a frequency-selective filtering framework aimed at suppressing shared frequency components between adjacent distinct actions while amplifying their action-specific frequencies, thereby enhancing inter-action discrepancies and sharpening transition boundaries. Specifically, Spectral Scalpel employs adaptive multi-scale spectral filters as scalpels to edit frequency spectra, coupled with a discrepancy loss between adjacent actions serving as the surgical objective. This design amplifies representational disparities between neighboring actions, effectively mitigating boundary localization ambiguities and inter-class confusion. Furthermore, complementing long-term temporal modeling, we introduce a frequency-aware channel mixer to strengthen channel evolution by aggregating spectra across channels. This work presents a novel paradigm for STAS that extends conventional spatio-temporal modeling by incorporating frequency-domain analysis. Extensive experiments on five public datasets demonstrate that Spectral Scalpel achieves state-of-the-art performance. Code is available at https://github.com/HaoyuJi/SpecScalpel.

Zhiyong Wang, Ge Zhang, Nineli Lashkarashvili

This paper describes our system for the SemEval2022 task of matching dictionary glosses to word embeddings. We focus on the Reverse Dictionary Track of the competition, which maps multilingual glosses to reconstructed vector representations. More specifically, models convert the input of sentences to three types of embeddings: SGNS, Char, and Electra. We propose several experiments for applying neural network cells, general multilingual and multitask structures, and language-agnostic tricks to the task. We also provide comparisons over different types of word embeddings and ablation studies to suggest helpful strategies. Our initial transformer-based model achieves relatively low performance. However, trials on different retokenization methodologies indicate improved performance. Our proposed Elmobased monolingual model achieves the highest outcome, and its multitask, and multilingual varieties show competitive results as well.

Penghui Wen, Kun Hu, Dong Yuan et al.

Radio frequency (RF) signals have been proved to be flexible for human silhouette segmentation (HSS) under complex environments. Existing studies are mainly based on a one-shot approach, which lacks a coherent projection ability from the RF domain. Additionally, the spatio-temporal patterns have not been fully explored for human motion dynamics in HSS. Therefore, we propose a two-stage Sequential Diffusion Model (SDM) to progressively synthesize high-quality segmentation jointly with the considerations on motion dynamics. Cross-view transformation blocks are devised to guide the diffusion model in a multi-scale manner for comprehensively characterizing human related patterns in an individual frame such as directional projection from signal planes. Moreover, spatio-temporal blocks are devised to fine-tune the frame-level model to incorporate spatio-temporal contexts and motion dynamics, enhancing the consistency of the segmentation maps. Comprehensive experiments on a public benchmark -- HIBER demonstrate the state-of-the-art performance of our method with an IoU 0.732. Our code is available at https://github.com/ph-w2000/SDM.

Peggy Tang, Kun Hu, Rui Yan et al.

Extractive text summarisation aims to select salient sentences from a document to form a short yet informative summary. While learning-based methods have achieved promising results, they have several limitations, such as dependence on expensive training and lack of interpretability. Therefore, in this paper, we propose a novel non-learning-based method by for the first time formulating text summarisation as an Optimal Transport (OT) problem, namely Optimal Transport Extractive Summariser (OTExtSum). Optimal sentence extraction is conceptualised as obtaining an optimal summary that minimises the transportation cost to a given document regarding their semantic distributions. Such a cost is defined by the Wasserstein distance and used to measure the summary's semantic coverage of the original document. Comprehensive experiments on four challenging and widely used datasets - MultiNews, PubMed, BillSum, and CNN/DM demonstrate that our proposed method outperforms the state-of-the-art non-learning-based methods and several recent learning-based methods in terms of the ROUGE metric.

Yuankun Xie, Chenxu Xiong, Xiaopeng Wang et al.

Currently, Audio Language Models (ALMs) are rapidly advancing due to the developments in large language models and audio neural codecs. These ALMs have significantly lowered the barrier to creating deepfake audio, generating highly realistic and diverse types of deepfake audio, which pose severe threats to society. Consequently, effective audio deepfake detection technologies to detect ALM-based audio have become increasingly critical. This paper investigate the effectiveness of current countermeasure (CM) against ALM-based audio. Specifically, we collect 12 types of the latest ALM-based deepfake audio and utilizing the latest CMs to evaluate. Our findings reveal that the latest codec-trained CM can effectively detect ALM-based audio, achieving 0% equal error rate under most ALM test conditions, which exceeded our expectations. This indicates promising directions for future research in ALM-based deepfake audio detection.

Yu Luo, Xiaogang Zhu, Shan Zeng et al.

Accurate crop yield prediction is crucial for sustainable agriculture and global food security. While existing methods are predominantly developed for single-crop prediction, they often struggle to generalize across diverse crop types, without addressing the unique crop phenological responses that are dynamically modulated by complex weather patterns. In this paper, we propose PhenoYieldNet, a multi-crop yield prediction framework that learns crop-specific phenology by explicitly modeling their responses with temporal drivers. Specifically, we develop a crop-aware temporal decoder consisting of a Crop Phenology Bank (CPB) and a Crop Phenology Attention (CPA) module. The CPB integrates a set of learnable embeddings, which leverage a query to guide the CPA module to learn the most relevant phenology patterns for the specific crop. And the CPA module explicitly captures multi-scale trend and variation components to construct temporal contexts, enabling the model to dynamically adjust the attention across different phenological stages. To learn robust and generalizable features for multi-crop prediction, the encoder is initialized with a pre-trained foundation model, and further adapted via a self-supervised Temporal Contrastive Adaptation strategy to align with agricultural temporal dynamics. Extensive experiments conducted on multi-crop datasets indicate that our proposed method significantly outperforms state-of-the-art methods, exhibiting strong generalization capabilities across different regions and crops.

Yuankun Xie, Xiaopeng Wang, Zhiyong Wang et al.

ASVspoof5, the fifth edition of the ASVspoof series, is one of the largest global audio security challenges. It aims to advance the development of countermeasure (CM) to discriminate bonafide and spoofed speech utterances. In this paper, we focus on addressing the problem of open-domain audio deepfake detection, which corresponds directly to the ASVspoof5 Track1 open condition. At first, we comprehensively investigate various CM on ASVspoof5, including data expansion, data augmentation, and self-supervised learning (SSL) features. Due to the high-frequency gaps characteristic of the ASVspoof5 dataset, we introduce Frequency Mask, a data augmentation method that masks specific frequency bands to improve CM robustness. Combining various scale of temporal information with multiple SSL features, our experiments achieved a minDCF of 0.0158 and an EER of 0.55% on the ASVspoof 5 Track 1 evaluation progress set.

Jianfeng Weng, Kun Hu, Tingting Yao et al.

Person Re-identification (ReID) has been extensively studied in recent years due to the increasing demand in public security. However, collecting and dealing with sensitive personal data raises privacy concerns. Therefore, federated learning has been explored for Person ReID, which aims to share minimal sensitive data between different parties (clients). However, existing federated learning based person ReID methods generally rely on laborious and time-consuming data annotations and it is difficult to guarantee cross-domain consistency. Thus, in this work, a federated unsupervised cluster-contrastive (FedUCC) learning method is proposed for Person ReID. FedUCC introduces a three-stage modelling strategy following a coarse-to-fine manner. In detail, generic knowledge, specialized knowledge and patch knowledge are discovered using a deep neural network. This enables the sharing of mutual knowledge among clients while retaining local domain-specific knowledge based on the kinds of network layers and their parameters. Comprehensive experiments on 8 public benchmark datasets demonstrate the state-of-the-art performance of our proposed method.

Zexin Hu, Kun Hu, Clinton Mo et al.

Sketch-based terrain generation seeks to create realistic landscapes for virtual environments in various applications such as computer games, animation and virtual reality. Recently, deep learning based terrain generation has emerged, notably the ones based on generative adversarial networks (GAN). However, these methods often struggle to fulfill the requirements of flexible user control and maintain generative diversity for realistic terrain. Therefore, we propose a novel diffusion-based method, namely terrain diffusion network (TDN), which actively incorporates user guidance for enhanced controllability, taking into account terrain features like rivers, ridges, basins, and peaks. Instead of adhering to a conventional monolithic denoising process, which often compromises the fidelity of terrain details or the alignment with user control, a multi-level denoising scheme is proposed to generate more realistic terrains by taking into account fine-grained details, particularly those related to climatic patterns influenced by erosion and tectonic activities. Specifically, three terrain synthesisers are designed for structural, intermediate, and fine-grained level denoising purposes, which allow each synthesiser concentrate on a distinct terrain aspect. Moreover, to maximise the efficiency of our TDN, we further introduce terrain and sketch latent spaces for the synthesizers with pre-trained terrain autoencoders. Comprehensive experiments on a new dataset constructed from NASA Topology Images clearly demonstrate the effectiveness of our proposed method, achieving the state-of-the-art performance. Our code and dataset will be publicly available.

Xin Qi, Ruibo Fu, Zhengqi Wen et al.

In recent years, speech diffusion models have advanced rapidly. Alongside the widely used U-Net architecture, transformer-based models such as the Diffusion Transformer (DiT) have also gained attention. However, current DiT speech models treat Mel spectrograms as general images, which overlooks the specific acoustic properties of speech. To address these limitations, we propose a method called Directional Patch Interaction for Text-to-Speech (DPI-TTS), which builds on DiT and achieves fast training without compromising accuracy. Notably, DPI-TTS employs a low-to-high frequency, frame-by-frame progressive inference approach that aligns more closely with acoustic properties, enhancing the naturalness of the generated speech. Additionally, we introduce a fine-grained style temporal modeling method that further improves speaker style similarity. Experimental results demonstrate that our method increases the training speed by nearly 2 times and significantly outperforms the baseline models.

Jiahang Liu, Tianyu Xu, Jiawei Chen et al.

Recent embodied navigation approaches leveraging Vision-Language Models (VLMs) demonstrate strong generalization in versatile Vision-Language Navigation (VLN). However, reliable path planning in complex environments remains challenging due to insufficient spatial awareness. In this work, we introduce SPAN-Nav, an end-to-end foundation model designed to infuse embodied navigation with universal 3D spatial awareness using RGB video streams. SPAN-Nav extracts spatial priors across diverse scenes through an occupancy prediction task on extensive indoor and outdoor environments. To mitigate the computational burden, we introduce a compact representation for spatial priors, finding that a single token is sufficient to encapsulate the coarse-grained cues essential for navigation tasks. Furthermore, inspired by the Chain-of-Thought (CoT) mechanism, SPAN-Nav utilizes this single spatial token to explicitly inject spatial cues into action reasoning through an end-to end framework. Leveraging multi-task co-training, SPAN-Nav captures task-adaptive cues from generalized spatial priors, enabling robust spatial awareness to generalize even to the task lacking explicit spatial supervision. To support comprehensive spatial learning, we present a massive dataset of 4.2 million occupancy annotations that covers both indoor and outdoor scenes across multi-type navigation tasks. SPAN-Nav achieves state-of-the-art performance across three benchmarks spanning diverse scenarios and varied navigation tasks. Finally, real-world experiments validate the robust generalization and practical reliability of our approach across complex physical scenarios.

Yuankun Xie, Yi Lu, Ruibo Fu et al.

With the proliferation of Audio Language Model (ALM) based deepfake audio, there is an urgent need for generalized detection methods. ALM-based deepfake audio currently exhibits widespread, high deception, and type versatility, posing a significant challenge to current audio deepfake detection (ADD) models trained solely on vocoded data. To effectively detect ALM-based deepfake audio, we focus on the mechanism of the ALM-based audio generation method, the conversion from neural codec to waveform. We initially constructed the Codecfake dataset, an open-source, large-scale collection comprising over 1 million audio samples in both English and Chinese, focus on ALM-based audio detection. As countermeasure, to achieve universal detection of deepfake audio and tackle domain ascent bias issue of original sharpness aware minimization (SAM), we propose the CSAM strategy to learn a domain balanced and generalized minima. In our experiments, we first demonstrate that ADD model training with the Codecfake dataset can effectively detects ALM-based audio. Furthermore, our proposed generalization countermeasure yields the lowest average equal error rate (EER) of 0.616% across all test conditions compared to baseline models. The dataset and associated code are available online.

Zhiyong Wang, Dongruo Zhou, John C. S. Lui et al.

Learning a transition model via Maximum Likelihood Estimation (MLE) followed by planning inside the learned model is perhaps the most standard and simplest Model-based Reinforcement Learning (RL) framework. In this work, we show that such a simple Model-based RL scheme, when equipped with optimistic and pessimistic planning procedures, achieves strong regret and sample complexity bounds in online and offline RL settings. Particularly, we demonstrate that under the conditions where the trajectory-wise reward is normalized between zero and one and the transition is time-homogenous, it achieves nearly horizon-free and second-order bounds. Nearly horizon-free means that our bounds have no polynomial dependence on the horizon of the Markov Decision Process. A second-order bound is a type of instance-dependent bound that scales with respect to the variances of the returns of the policies which can be small when the system is nearly deterministic and (or) the optimal policy has small values. We highlight that our algorithms are simple, fairly standard, and indeed have been extensively studied in the RL literature: they learn a model via MLE, build a version space around the MLE solution, and perform optimistic or pessimistic planning depending on whether operating in the online or offline mode. These algorithms do not rely on additional specialized algorithmic designs such as learning variances and performing variance-weighted learning and thus can easily leverage non-linear function approximations. The simplicity of the algorithms also implies that our horizon-free and second-order regret analysis is actually standard and mainly follows the general framework of optimism/pessimism in the face of uncertainty.

Haoran He, Yuxiao Ye, Jie Liu et al.

Fine-tuning diffusion models via online reinforcement learning (RL) has shown great potential for enhancing text-to-image alignment. However, since precisely specifying a ground-truth objective for visual tasks remains challenging, the models are often optimized using a proxy reward that only partially captures the true goal. This mismatch often leads to reward hacking, where proxy scores increase while real image quality deteriorates and generation diversity collapses. While common solutions add regularization against the reference policy to prevent reward hacking, they compromise sample efficiency and impede the exploration of novel, high-reward regions, as the reference policy is usually sub-optimal. To address the competing demands of sample efficiency, effective exploration, and mitigation of reward hacking, we propose Gated and Adaptive Regularization with Diversity-aware Optimization (GARDO), a versatile framework compatible with various RL algorithms. Our key insight is that regularization need not be applied universally; instead, it is highly effective to selectively penalize a subset of samples that exhibit high uncertainty. To address the exploration challenge, GARDO introduces an adaptive regularization mechanism wherein the reference model is periodically updated to match the capabilities of the online policy, ensuring a relevant regularization target. To address the mode collapse issue in RL, GARDO amplifies the rewards for high-quality samples that also exhibit high diversity, encouraging mode coverage without destabilizing the optimization process. Extensive experiments across diverse proxy rewards and hold-out unseen metrics consistently show that GARDO mitigates reward hacking and enhances generation diversity without sacrificing sample efficiency or exploration, highlighting its effectiveness and robustness.

Peggy Tang, Junbin Gao, Lei Zhang et al.

Recently, compressive text summarisation offers a balance between the conciseness issue of extractive summarisation and the factual hallucination issue of abstractive summarisation. However, most existing compressive summarisation methods are supervised, relying on the expensive effort of creating a new training dataset with corresponding compressive summaries. In this paper, we propose an efficient and interpretable compressive summarisation method that utilises unsupervised dual-agent reinforcement learning to optimise a summary's semantic coverage and fluency by simulating human judgment on summarisation quality. Our model consists of an extractor agent and a compressor agent, and both agents have a multi-head attentional pointer-based structure. The extractor agent first chooses salient sentences from a document, and then the compressor agent compresses these extracted sentences by selecting salient words to form a summary without using reference summaries to compute the summary reward. To our best knowledge, this is the first work on unsupervised compressive summarisation. Experimental results on three widely used datasets (e.g., Newsroom, CNN/DM, and XSum) show that our model achieves promising performance and a significant improvement on Newsroom in terms of the ROUGE metric, as well as interpretability of semantic coverage of summarisation results.

Baoli Sun, Yihan Wang, Xinzhu Ma et al.

Fine-grained action recognition (FGAR) aims to identify subtle and distinctive differences among fine-grained action categories. However, current recognition methods often capture coarse-grained motion patterns but struggle to identify subtle details in local regions evolving over time. In this work, we introduce the Action-Region Tracking (ART) framework, a novel solution leveraging a query-response mechanism to discover and track the dynamics of distinctive local details, enabling effective distinction of similar actions. Specifically, we propose a region-specific semantic activation module that employs discriminative and text-constrained semantics as queries to capture the most action-related region responses in each video frame, facilitating interaction among spatial and temporal dimensions with corresponding video features. The captured region responses are organized into action tracklets, which characterize region-based action dynamics by linking related responses across video frames in a coherent sequence. The text-constrained queries encode nuanced semantic representations derived from textual descriptions of action labels extracted by language branches within Visual Language Models (VLMs). To optimize the action tracklets, we design a multi-level tracklet contrastive constraint among region responses at spatial and temporal levels, enabling effective discrimination within each frame and correlation between adjacent frames. Additionally, a task-specific fine-tuning mechanism refines textual semantics such that semantic representations encoded by VLMs are preserved while optimized for task preferences. Comprehensive experiments on widely used action recognition benchmarks demonstrate the superiority to previous state-of-the-art baselines.

Baoli Sun, Xinzhu Ma, Ning Wang et al.

Referring video object segmentation (RVOS) is an emerging cross-modality task that aims to generate pixel-level maps of the target objects referred by given textual expressions. The main concept involves learning an accurate alignment of visual elements and language expressions within a semantic space. Recent approaches address cross-modality alignment through conditional queries, tracking the target object using a query-response based mechanism built upon transformer structure. However, they exhibit two limitations: (1) these conditional queries lack inter-frame dependency and variation modeling, making accurate target tracking challenging amid significant frame-to-frame variations; and (2) they integrate textual constraints belatedly, which may cause the video features potentially focus on the non-referred objects. Therefore, we propose a novel RVOS architecture called ProxyFormer, which introduces a set of proxy queries to integrate visual and text semantics and facilitate the flow of semantics between them. By progressively updating and propagating proxy queries across multiple stages of video feature encoder, ProxyFormer ensures that the video features are focused on the object of interest. This dynamic evolution also enables the establishment of inter-frame dependencies, enhancing the accuracy and coherence of object tracking. To mitigate high computational costs, we decouple cross-modality interactions into temporal and spatial dimensions. Additionally, we design a Joint Semantic Consistency (JSC) training strategy to align semantic consensus between the proxy queries and the combined video-text pairs. Comprehensive experiments on four widely used RVOS benchmarks demonstrate the superiority of our ProxyFormer to the state-of-the-art methods.

Wenxi Yue, Jing Zhang, Kun Hu et al.

The Segment Anything Model (SAM) exhibits promise in generic object segmentation and offers potential for various applications. Existing methods have applied SAM to surgical instrument segmentation (SIS) by tuning SAM-based frameworks with surgical data. However, they fall short in two crucial aspects: (1) Straightforward model tuning with instrument masks treats each instrument as a single entity, neglecting their complex structures and fine-grained details; and (2) Instrument category-based prompts are not flexible and informative enough to describe instrument structures. To address these problems, in this paper, we investigate text promptable SIS and propose SurgicalPart-SAM (SP-SAM), a novel SAM efficient-tuning approach that explicitly integrates instrument structure knowledge with SAM's generic knowledge, guided by expert knowledge on instrument part compositions. Specifically, we achieve this by proposing (1) Collaborative Prompts that describe instrument structures via collaborating category-level and part-level texts; (2) Cross-Modal Prompt Encoder that encodes text prompts jointly with visual embeddings into discriminative part-level representations; and (3) Part-to-Whole Adaptive Fusion and Hierarchical Decoding that adaptively fuse the part-level representations into a whole for accurate instrument segmentation in surgical scenarios. Built upon them, SP-SAM acquires a better capability to comprehend surgical instruments in terms of both overall structure and part-level details. Extensive experiments on both the EndoVis2018 and EndoVis2017 datasets demonstrate SP-SAM's state-of-the-art performance with minimal tunable parameters. The code will be available at https://github.com/wenxi-yue/SurgicalPart-SAM.

Penghui Wen, Zexin Hu, Sen Zhang et al.

Precipitation nowcasting remains challenging due to the highly localized, rapidly evolving, and heterogeneous nature of atmospheric dynamics. Although recent methods increasingly adopt attention-based architectures in both unimodal and multimodal settings, they mainly emphasize stronger representation learning and prediction capacity, while paying less attention to the stability of attention responses across samples. In this work, we show that cross-sample instability of attention-response energy is an important and previously underexplored source of forecasting unreliability. Empirically, inaccurate forecasts are associated with larger attention-response energy variance across heads and layers. Theoretically, we show that cross-sample variability can propagate through self-attention, and enlarge a lower bound on prediction error. Based on this insight, we propose HARECast, a Head-wise Attention Response Energy-regulated framework for precipitation nowcasting. HARECast explicitly models head-wise attention-response energy and stabilizes it through a group-wise regularization objective that reduces cross-sample fluctuations. The proposed formulation is generic and applicable to both unimodal and multimodal nowcasting architectures. We instantiate HARECast in a standard forecasting pipeline with reconstruction branches and a diffusion-based predictor, and evaluate it on commonly used benchmarks--SEVIR and MeteoNet. Experimental results demonstrate that HARECast achieves state-of-the-art performance.

Penghui Wen, Yu Luo, Lintao Wang et al.

Existing precipitation nowcasting methods typically adopt an autoregressive formulation, where future states are predicted from previous outputs. However, such an approach accumulates errors over long rollouts, causing forecasts to drift away from physically plausible evolution trajectories. Although various studies have attempted to alleviate this problem by improving step-wise prediction accuracy, they largely neglect the global temporal evolution of meteorological systems and lack mechanisms to actively correct drift during rollouts. To address this issue, we propose McCast, a memory-guided latent drift correction method for precipitation nowcasting. Rather than treating memory as an unordered dictionary of latent states for passive conditioning, McCast leverages temporally organized memory to actively correct autoregressive latent evolution. Specifically, McCast introduces a Drift-Corrective Memory Bank (DCBank) that explicitly estimates the temporally consistent drift corrections to calibrate the divergent trajectory. DCBank performs drift correction in two stages: a Corrective Latent Extractor first predicts an initial correction from the current prediction and a reference latent state, and a Correction-Aware Memory Retrieval module then refines the initial correction using temporally organized historical memory. By explicitly correcting latent evolution, instead of improving step-wise prediction accuracy only, McCast produces more temporally coherent and reliable long-horizon forecasts. Experiments on two widely used benchmarks, SEVIR and MeteoNet, show that McCast achieves state-of-the-art performance, particularly in challenging long-horizon forecasting scenarios.

Zhuqiang Lu, Zhenfei Yin, Mengwei He et al.

Recently, Vision Large Language Models (VLLMs) integrated with vision encoders have shown promising performance in vision understanding. The key of VLLMs is to encode visual content into sequences of visual tokens, enabling VLLMs to simultaneously process both visual and textual content. However, understanding videos, especially long videos, remain a challenge to VLLMs as the number of visual tokens grows rapidly when encoding videos, resulting in the risk of exceeding the context window of VLLMs and introducing heavy computation burden. To restrict the number of visual tokens, existing VLLMs either: (1) uniformly downsample videos into a fixed number of frames or (2) reducing the number of visual tokens encoded from each frame. We argue the former solution neglects the rich temporal cue in videos and the later overlooks the spatial details in each frame. In this work, we present Balanced-VLLM (B-VLLM): a novel VLLM framework that aims to effectively leverage task relevant spatio-temporal cues while restricting the number of visual tokens under the VLLM context window length. At the core of our method, we devise a text-conditioned adaptive frame selection module to identify frames relevant to the visual understanding task. The selected frames are then de-duplicated using a temporal frame token merging technique. The visual tokens of the selected frames are processed through a spatial token sampling module and an optional spatial token merging strategy to achieve precise control over the token count. Experimental results show that B-VLLM is effective in balancing the number of frames and visual tokens in video understanding, yielding superior performance on various video understanding benchmarks. Our code is available at https://github.com/zhuqiangLu/B-VLLM.

Chengbing Chen, Yutong Li, Rui Zhao et al.

As a machine-learned potential, the neuroevolution potential (NEP) method features exceptional computational efficiency and has been successfully applied in materials science. Constructing high-quality training datasets is crucial for developing accurate NEP models. However, the preparation and screening of NEP training datasets remain a bottleneck for broader applications due to their time-consuming, labor-intensive, and resource-intensive nature. In this work, we have developed NepTrain and NepTrainKit, which are dedicated to initializing and managing training datasets to generate high-quality training sets while automating NEP model training. NepTrain is an open-source Python package that features a bond length filtering method to effectively identify and remove non-physical structures from molecular dynamics trajectories, thereby ensuring high-quality training datasets. NepTrainKit is a graphical user interface (GUI) software designed specifically for NEP training datasets, providing functionalities for data editing, visualization, and interactive exploration. It integrates key features such as outlier identification, farthest-point sampling, non-physical structure detection, and configuration type selection. The combination of these tools enables users to process datasets more efficiently and conveniently. Using $\rm CsPbI_3$ as a case study, we demonstrate the complete workflow for training NEP models with NepTrain and further validate the models through materials property predictions. We believe this toolkit will greatly benefit researchers working with machine learning interatomic potentials.

Yuankun Xie, Ruibo Fu, Zhiyong Wang et al.

The rapid advancement of audio generation technologies has escalated the risks of malicious deepfake audio across speech, sound, singing voice, and music, threatening multimedia security and trust. While existing countermeasures (CMs) perform well in single-type audio deepfake detection (ADD), their performance declines in cross-type scenarios. This paper is dedicated to studying the alltype ADD task. We are the first to comprehensively establish an all-type ADD benchmark to evaluate current CMs, incorporating cross-type deepfake detection across speech, sound, singing voice, and music. Then, we introduce the prompt tuning self-supervised learning (PT-SSL) training paradigm, which optimizes SSL frontend by learning specialized prompt tokens for ADD, requiring 458x fewer trainable parameters than fine-tuning (FT). Considering the auditory perception of different audio types,we propose the wavelet prompt tuning (WPT)-SSL method to capture type-invariant auditory deepfake information from the frequency domain without requiring additional training parameters, thereby enhancing performance over FT in the all-type ADD task. To achieve an universally CM, we utilize all types of deepfake audio for co-training. Experimental results demonstrate that WPT-XLSR-AASIST achieved the best performance, with an average EER of 3.58% across all evaluation sets. The code is available online.

Yuchen Huang, Baiteng Ma, Yiping Sun et al.

Vector approximate nearest neighbor search (ANNS) underpins search engines, recommendation systems, and advertising services. Recent advances in ANNS indexes make CPU a cost-effective choice for serving million-scale, in-memory vector search, yet per-core throughput remains constrained by memory access latency of vector reading and the compute intensity of distance evaluations in production deployments. With the growing scale of the business and advances in hardware, modern CCD-based multi-core CPUs have been widely deployed for high throughput in our services. However, we find that simply increasing core counts does not yield optimal performance scaling. To improve the efficiency of more cores from the CCD-based architecture, we analyze the distributions of real-world requests in our production environments. We observe high access locality in vector search in our online services and low cache utilization, resulting from overlooking the multi-chiplet nature of CCD based CPUs. Hence, we propose a workload- and hardware-aware thread orchestration framework at CCD-level that (i) provides a uniform interface for both inter-query parallel HNSW search and intra-query parallel IVF search, (ii) achieves cache-friendly and workload-adaptive mapping of task dispatching, and (iii) employs CCD-aware task stealing to address load imbalance. Applied to real production workloads from search, recommendation, and advertising services of Xiaohongshu (RedNote), our approach delivers up to 3.7x higher throughput and 30-90% reductions in P50 and P999 latency. In detail, compared with the original framework, the cache-miss ratio decreases by 6-30%, and the total CPU stall is reduced by 20-80%.

Ruhan Wang, Chengkai Huang, Zhiyong Wang et al.

Large language models (LLMs) exhibit strong reasoning capabilities when guided by high-quality demonstrations, yet such data is often distributed across organizations that cannot centralize it due to regulatory, proprietary, or institutional constraints. We study federated reasoning, where a server improves multi-step reasoning by coordinating with heterogeneous clients holding private demonstrations, without centralized training or raw data sharing. The key challenge is that client reliability is query-dependent, while the server cannot inspect client data to determine which contributions are trustworthy. To address this, we propose Uncertainty-Aware Federated Reasoning (FERA), a training-free framework based on iterative server-client co-refinement. Across communication rounds, clients generate reasoning traces with lightweight uncertainty estimates, and the server synthesizes them into improved reasoning that is redistributed as context for the next round, progressively improving both server outputs and client-side reasoning. Within each round, Uncertainty-Aware Self-Critique Aggregation (UA-SCA) resolves conflicts among heterogeneous client traces through query-dependent trust weighting and structured cross-client verification. Rather than simply discarding low-quality traces, UA-SCA revises flawed reasoning steps to recover useful information. We provide theoretical guarantees showing that the proposed iterative protocol converges and that uncertainty-aware weighting accelerates convergence. Experiments on multiple reasoning benchmarks show that FERA consistently outperforms both federated training and training-free baselines, achieving progressively higher accuracy across rounds while maintaining communication and computational efficiency.