Xun Gong

Ziqiang Zhang, Sanyuan Chen, Long Zhou et al. · microsoft-research

How to boost speech pre-training with textual data is an unsolved problem due to the fact that speech and text are very different modalities with distinct characteristics. In this paper, we propose a cross-modal Speech and Language Model (SpeechLM) to explicitly align speech and text pre-training with a pre-defined unified discrete representation. Specifically, we introduce two alternative discrete tokenizers to bridge the speech and text modalities, including phoneme-unit and hidden-unit tokenizers, which can be trained using a small amount of paired speech-text data. Based on the trained tokenizers, we convert the unlabeled speech and text data into tokens of phoneme units or hidden units. The pre-training objective is designed to unify the speech and the text into the same discrete semantic space with a unified Transformer network. We evaluate SpeechLM on various spoken language processing tasks including speech recognition, speech translation, and universal representation evaluation framework SUPERB, demonstrating significant improvements on content-related tasks. Code and models are available at https://aka.ms/SpeechLM.

Mohammad Reza Amini, Xun Gong, Yiheng Feng et al.

The emergence of connected and automated vehicles (CAVs) provides an unprecedented opportunity to capitalize on these technologies well beyond their original designed intents. While abundant evidence has been accumulated showing substantial fuel economy improvement benefits achieved through advanced powertrain control, the implications of the CAV operation on power and thermal management have not been fully investigated. In this paper, in order to explore the opportunities for the coordination between the onboard thermal management and the power split control, we present a sequential optimization solution for eco-driving speed trajectory planning, air conditioning (A/C) thermal load planning (eco-cooling), and powertrain control in hybrid electric CAVs to evaluate the individual as well as the collective energy savings through proactive usage of traffic data for vehicle speed prediction. Simulation results over a real-world driving cycle show that compared to a baseline non-CAV, 11.9%, 14.2%, and 18.8% energy savings can be accumulated sequentially through speed, thermal load, and power split optimizations, respectively.

Xun Gong, Yu Wu, Jinyu Li et al.

Traditional automatic speech recognition~(ASR) systems usually focus on individual utterances, without considering long-form speech with useful historical information, which is more practical in real scenarios. Simply attending longer transcription history for a vanilla neural transducer model shows no much gain in our preliminary experiments, since the prediction network is not a pure language model. This motivates us to leverage the factorized neural transducer structure, containing a real language model, the vocabulary predictor. We propose the {LongFNT-Text} architecture, which fuses the sentence-level long-form features directly with the output of the vocabulary predictor and then embeds token-level long-form features inside the vocabulary predictor, with a pre-trained contextual encoder RoBERTa to further boost the performance. Moreover, we propose the {LongFNT} architecture by extending the long-form speech to the original speech input and achieve the best performance. The effectiveness of our LongFNT approach is validated on LibriSpeech and GigaSpeech corpora with 19% and 12% relative word error rate~(WER) reduction, respectively.

Yuankun Chen, Zifei Nie, Xun Gong et al.

Differentiable optimal control, particularly differentiable nonlinear model predictive control (NMPC), provides a powerful framework that enjoys the complementary benefits of machine learning and control theory. A key enabler of differentiable optimal control is the computation of derivatives of the optimal trajectory with respect to problem parameters, i.e., trajectory derivatives. Previous works compute trajectory derivatives by solving a differential Karush-Kuhn-Tucker (KKT) system, and achieve this efficiently by constructing an equivalent auxiliary system. However, we find that directly exploiting the matrix structures in the differential KKT system yields significant computation speed improvements. Motivated by this insight, we propose FastDOC, which applies a Gauss-Newton approximation of Hessian and takes advantage of the resulting block-sparsity and positive semidefinite properties of the matrices involved. These structural properties enable us to accelerate the computationally expensive matrix factorization steps, resulting in a factor-of-two speedup in theoretical computational complexity, and in a synthetic benchmark FastDOC achieves up to a 180% time reduction compared to the baseline method. Finally, we validate the method on an imitation learning task for human-like autonomous driving, where the results demonstrate the effectiveness of the proposed FastDOC in practical applications.

Shijie Wu, Xun Gong

Face recognition has made tremendous progress in recent years due to the advances in loss functions and the explosive growth in training sets size. A properly designed loss is seen as key to extract discriminative features for classification. Several margin-based losses have been proposed as alternatives of softmax loss in face recognition. However, two issues remain to consider: 1) They overlook the importance of hard sample mining for discriminative learning. 2) Label noise ubiquitously exists in large-scale datasets, which can seriously damage the model's performance. In this paper, starting from the perspective of decision boundary, we propose a novel mining framework that focuses on the relationship between a sample's ground truth class center and its nearest negative class center. Specifically, a closed-set noise label self-correction module is put forward, making this framework work well on datasets containing a lot of label noise. The proposed method consistently outperforms SOTA methods in various face recognition benchmarks. Training code has been released at https://github.com/SWJTU-3DVision/BoundaryFace.

Xun Gong, Yaohui Guo, Yiheng Feng et al.

The energy efficiency of Connected and Automated Vehicles (CAVs) is significantly influenced by surrounding road users. This paper presents the evaluation of energy efficiency of CAVs in a mixed traffic interacted with human controlled vehicles. To simulate the interaction between the CAVs and the cut-in vehicles controlled by human drivers near the intersection, a lane changing model is proposed to emulate the politeness and patience characteristics of the human driver. The proposed lane changing model is then calibrated based on over 100,000 naturalistic lane changing events collected by the University of Michigan Safety Pilot Model Deployment Program. A case study on simulation of the cut-in scenario is carried out to demonstrate the human driver's lane changing sensitivity under different driving trajectories of a frontal CAV and the influence on the energy consumption of the CAV due to the cut-in vehicle is evaluated. The simulation results indicate that the fuel economy of the CAV can be substantially improved if its surrounding cut-in vehicles can be well handled.

Lin Fan, Yafei Ou, Zhipeng Deng et al.

Chain-of-thought (CoT) reasoning has advanced medical visual question answering (VQA), yet most existing CoT rationales are free-form and fail to capture the structured reasoning process clinicians actually follow. This work asks: Can traceable, multi-step reasoning supervision improve reasoning accuracy and the interpretability of Medical VQA? To this end, we introduce Step-CoT, a large-scale medical reasoning dataset with expert-curated, structured multi-step CoT aligned to clinical diagnostic workflows, implicitly grounding the model's reasoning in radiographic evidence. Step-CoT comprises more than 10K real clinical cases and 70K VQA pairs organized around diagnostic workflows, providing supervised intermediate steps that guide models to follow valid reasoning trajectories. To effectively learn from Step-CoT, we further introduce a teacher-student framework with a dynamic graph-structured focusing mechanism that prioritizes diagnostically informative steps while filtering out less relevant contexts. Our experiments show that using Step-CoT can improve reasoning accuracy and interpretability. Benchmark: github.com/hahaha111111/Step-CoT. Dataset Card: huggingface.co/datasets/fl-15o/Step-CoT

Ping Chen, Xiang Liu, Xingpeng Zhang et al.

Diffusion models operate in a reflexive System 1 mode, constrained by a fixed, content-agnostic sampling schedule. This rigidity arises from the curse of state dimensionality, where the combinatorial explosion of possible states in the high-dimensional noise manifold renders explicit trajectory planning intractable and leads to systematic computational misallocation. To address this, we introduce Chain-of-Trajectories (CoTj), a train-free framework enabling System 2 deliberative planning. Central to CoTj is Diffusion DNA, a low-dimensional signature that quantifies per-stage denoising difficulty and serves as a proxy for the high-dimensional state space, allowing us to reformulate sampling as graph planning on a directed acyclic graph. Through a Predict-Plan-Execute paradigm, CoTj dynamically allocates computational effort to the most challenging generative phases. Experiments across multiple generative models demonstrate that CoTj discovers context-aware trajectories, improving output quality and stability while reducing redundant computation. This work establishes a new foundation for resource-aware, planning-based diffusion modeling. The code is available at https://github.com/UnicomAI/CoTj.

Zhemin Zhang, Xun Gong

Foundation models, such as CNNs and ViTs, have powered the development of image representation learning. However, general guidance to model architecture design is still missing. Inspired by the connection between image representation learning and heat conduction, we model images by the heat conduction equation, where the essential idea is to conceptualize image features as temperatures and model their information interaction as the diffusion of thermal energy. Based on this idea, we find that many modern model architectures, such as residual structures, SE block, and feed-forward networks, can be interpreted from the perspective of the heat conduction equation. Therefore, we leverage the heat equation to design new and more interpretable models. As an example, we propose the Heat Conduction Layer and the Refinement Approximation Layer inspired by solving the heat conduction equation using Finite Difference Method and Fourier series, respectively. The main goal of this paper is to integrate the overall architectural design of neural networks into the theoretical framework of heat conduction. Nevertheless, our Heat Conduction Network (HcNet) still shows competitive performance, e.g., HcNet-T achieves 83.0% top-1 accuracy on ImageNet-1K while only requiring 28M parameters and 4.1G MACs. The code is publicly available at: https://github.com/ZheminZhang1/HcNet.

Zhen Liu, Hang Gao, Hao Ma et al.

Autonomous vehicle (AV) evaluation has been the subject of increased interest in recent years both in industry and in academia. This paper focuses on the development of a novel framework for generating adversarial driving behavior of background vehicle interfering against the AV to expose effective and rational risky events. Specifically, the adversarial behavior is learned by a reinforcement learning (RL) approach incorporated with the cumulative prospect theory (CPT) which allows representation of human risk cognition. Then, the extended version of deep deterministic policy gradient (DDPG) technique is proposed for training the adversarial policy while ensuring training stability as the CPT action-value function is leveraged. A comparative case study regarding the cut-in scenario is conducted on a high fidelity Hardware-in-the-Loop (HiL) platform and the results demonstrate the adversarial effectiveness to infer the weakness of the tested AV.

Jinchuan Tian, Haoran Wang, Bo-Hao Su et al.

Current audio foundation models typically rely on rigid, task-specific supervision, addressing isolated factors of audio rather than the whole. In contrast, human intelligence processes audio holistically, seamlessly bridging physical signals with abstract cognitive concepts to execute complex tasks. Grounded in this philosophy, we introduce Bagpiper, an 8B audio foundation model that interprets physical audio via rich captions, i.e., comprehensive natural language descriptions that encapsulate the critical cognitive concepts inherent in the signal (e.g., transcription, audio events). By pre-training on a massive corpus of 600B tokens, the model establishes a robust bidirectional mapping between raw audio and this high-level conceptual space. During fine-tuning, Bagpiper adopts a caption-then-process workflow, simulating an intermediate cognitive reasoning step to solve diverse tasks without task-specific priors. Experimentally, Bagpiper outperforms Qwen-2.5-Omni on MMAU and AIRBench for audio understanding and surpasses CosyVoice3 and TangoFlux in generation quality, capable of synthesizing arbitrary compositions of speech, music, and sound effects. To the best of our knowledge, Bagpiper is among the first works that achieve unified understanding generation for general audio. Model, data, and code are available at Bagpiper Home Page.

Xuan Tan, Xun Gong, Yang Xiang

Contrastive Language-Image Pre-Training (CLIP) model excels in traditional person re-identification (ReID) tasks due to its inherent advantage in generating textual descriptions for pedestrian images. However, applying CLIP directly to intra-camera supervised person re-identification (ICS ReID) presents challenges. ICS ReID requires independent identity labeling within each camera, without associations across cameras. This limits the effectiveness of text-based enhancements. To address this, we propose a novel framework called CLIP-based Camera-Agnostic Feature Learning (CCAFL) for ICS ReID. Accordingly, two custom modules are designed to guide the model to actively learn camera-agnostic pedestrian features: Intra-Camera Discriminative Learning (ICDL) and Inter-Camera Adversarial Learning (ICAL). Specifically, we first establish learnable textual prompts for intra-camera pedestrian images to obtain crucial semantic supervision signals for subsequent intra- and inter-camera learning. Then, we design ICDL to increase inter-class variation by considering the hard positive and hard negative samples within each camera, thereby learning intra-camera finer-grained pedestrian features. Additionally, we propose ICAL to reduce inter-camera pedestrian feature discrepancies by penalizing the model's ability to predict the camera from which a pedestrian image originates, thus enhancing the model's capability to recognize pedestrians from different viewpoints. Extensive experiments on popular ReID datasets demonstrate the effectiveness of our approach. Especially, on the challenging MSMT17 dataset, we arrive at 58.9\% in terms of mAP accuracy, surpassing state-of-the-art methods by 7.6\%. Code will be available at: https://github.com/Trangle12/CCAFL.

Peng Huang, Shu Hu, Bo Peng et al.

MedSAM, a medical foundation model derived from the SAM architecture, has demonstrated notable success across diverse medical domains. However, its clinical application faces two major challenges: the dependency on labor-intensive manual prompt generation, which imposes a significant burden on clinicians, and the absence of semantic labeling in the generated segmentation masks for organs or lesions, limiting its practicality for non-expert users. To address these limitations, we propose AutoMedSAM, an end-to-end framework derived from SAM, designed to enhance usability and segmentation performance. AutoMedSAM retains MedSAM's image encoder and mask decoder structure while introducing a novel diffusion-based class prompt encoder. The diffusion-based encoder employs a dual-decoder structure to collaboratively generate prompt embeddings guided by sparse and dense prompt definitions. These embeddings enhance the model's ability to understand and process clinical imagery autonomously. With this encoder, AutoMedSAM leverages class prompts to embed semantic information into the model's predictions, transforming MedSAM's semi-automated pipeline into a fully automated workflow. Furthermore, AutoMedSAM employs an uncertainty-aware joint optimization strategy during training to effectively inherit MedSAM's pre-trained knowledge while improving generalization by integrating multiple loss functions. Experimental results across diverse datasets demonstrate that AutoMedSAM achieves superior performance while broadening its applicability to both clinical settings and non-expert users. Code is available at https://github.com/HP-ML/AutoPromptMedSAM.git.

Yanjiao Liu, Jiawei Liu, Xun Gong et al.

Large language models (LLMs) have recently demonstrated strong reasoning capabilities and attracted increasing research attention in the field of autonomous driving (AD). However, safe application of LLMs on AD perception and prediction still requires a thorough understanding of both the dynamic traffic agents and the static road infrastructure. To this end, this study introduces a framework to evaluate the capability of LLMs in understanding the behaviors of dynamic traffic agents and the topology of road networks. The framework leverages frozen LLMs as the reasoning engine, employing a traffic encoder to extract spatial-level scene features from observed trajectories of agents, while a lightweight Convolutional Neural Network (CNN) encodes the local high-definition (HD) maps. To assess the intrinsic reasoning ability of LLMs, the extracted scene features are then transformed into LLM-compatible tokens via a reprogramming adapter. By residing the prediction burden with the LLMs, a simpler linear decoder is applied to output future trajectories. The framework enables a quantitative analysis of the influence of multi-modal information, especially the impact of map semantics on trajectory prediction accuracy, and allows seamless integration of frozen LLMs with minimal adaptation, thereby demonstrating strong generalizability across diverse LLM architectures and providing a unified platform for model evaluation.

Zhemin Zhang, Xun Gong

Positional encoding is important for vision transformer (ViT) to capture the spatial structure of the input image. General effectiveness has been proven in ViT. In our work we propose to train ViT to recognize the positional label of patches of the input image, this apparently simple task actually yields a meaningful self-supervisory task. Based on previous work on ViT positional encoding, we propose two positional labels dedicated to 2D images including absolute position and relative position. Our positional labels can be easily plugged into various current ViT variants. It can work in two ways: (a) As an auxiliary training target for vanilla ViT (e.g., ViT-B and Swin-B) for better performance. (b) Combine the self-supervised ViT (e.g., MAE) to provide a more powerful self-supervised signal for semantic feature learning. Experiments demonstrate that with the proposed self-supervised methods, ViT-B and Swin-B gain improvements of 1.20% (top-1 Acc) and 0.74% (top-1 Acc) on ImageNet, respectively, and 6.15% and 1.14% improvement on Mini-ImageNet.

Qing Li, Huifang Feng, Xun Gong et al.

3D Gaussian Splatting has recently emerged as an efficient solution for high-quality and real-time novel view synthesis. However, its capability for accurate surface reconstruction remains underexplored. Due to the discrete and unstructured nature of Gaussians, supervision based solely on image rendering loss often leads to inaccurate geometry and inconsistent multi-view alignment. In this work, we propose a novel method that enhances the geometric representation of 3D Gaussians through view alignment (VA). Specifically, we incorporate edge-aware image cues into the rendering loss to improve surface boundary delineation. To enforce geometric consistency across views, we introduce a visibility-aware photometric alignment loss that models occlusions and encourages accurate spatial relationships among Gaussians. To further mitigate ambiguities caused by lighting variations, we incorporate normal-based constraints to refine the spatial orientation of Gaussians and improve local surface estimation. Additionally, we leverage deep image feature embeddings to enforce cross-view consistency, enhancing the robustness of the learned geometry under varying viewpoints and illumination. Extensive experiments on standard benchmarks demonstrate that our method achieves state-of-the-art performance in both surface reconstruction and novel view synthesis. The source code is available at https://github.com/LeoQLi/VA-GS.

Zhemin Zhang, Xun Gong

The linear classifier is widely used in various image classification tasks. It works by optimizing the distance between a sample and its corresponding class center. However, in real-world data, one class can contain several local clusters, e.g., birds of different poses. To address this complexity, we propose a novel multi-center classifier. Different from the vanilla linear classifier, our proposal is established on the assumption that the deep features of the training set follow a Gaussian Mixture distribution. Specifically, we create a conditional Gaussian distribution for each class and then sample multiple sub-centers from that distribution to extend the linear classifier. This approach allows the model to capture intra-class local structures more efficiently. In addition, at test time we set the mean of the conditional Gaussian distribution as the class center of the linear classifier and follow the vanilla linear classifier outputs, thus requiring no additional parameters or computational overhead. Extensive experiments on image classification show that the proposed multi-center classifier is a powerful alternative to widely used linear classifiers. Code available at https://github.com/ZheminZhang1/MultiCenter-Classifier.

Lin Fan, Pengyu Dai, Zhipeng Deng et al.

Clinical image interpretation is inherently multi-step and tool-centric: clinicians iteratively combine visual evidence with patient context, quantify findings, and refine their decisions through a sequence of specialized procedures. While LLM-based agents promise to orchestrate such heterogeneous medical tools, existing systems treat tool sets and invocation strategies as static after deployment. This design is brittle under real-world domain shifts, across tasks, and evolving diagnostic requirements, where predefined tool chains frequently degrade and demand costly manual re-design. We propose MACRO, a self-evolving, experience-augmented medical agent that shifts from static tool composition to experience-driven tool discovery. From verified execution trajectories, the agent autonomously identifies recurring effective multi-step tool sequences, synthesizes them into reusable composite tools, and registers these as new high-level primitives that continuously expand its behavioral repertoire. A lightweight image-feature memory grounds tool selection in a visual-clinical context, while a GRPO-like training loop reinforces reliable invocation of discovered composites, enabling closed-loop self-improvement with minimal supervision. Extensive experiments across diverse medical imaging datasets and tasks demonstrate that autonomous composite tool discovery consistently improves multi-step orchestration accuracy and cross-domain generalization over strong baselines and recent state-of-the-art agentic methods, bridging the gap between brittle static tool use and adaptive, context-aware clinical AI assistance. Code will be available upon acceptance.

Zhemin Zhang, Xun Gong

Recently, Transformers have shown promising performance in various vision tasks. However, the high costs of global self-attention remain challenging for Transformers, especially for high-resolution vision tasks. Inspired by one of the most successful transformers-based models for NLP: Big Bird, we propose a novel sparse attention mechanism for Vision Transformers (ViT). Specifically, we separate the heads into three groups, the first group used convolutional neural network (CNN) to extract local features and provide positional information for the model, the second group used Random Sampling Windows (RS-Win) for sparse self-attention calculation, and the third group reduces the resolution of the keys and values by average pooling for global attention. Based on these components, ViT maintains the sparsity of self-attention while maintaining the merits of Big Bird (i.e., the model is a universal approximator of sequence functions and is Turing complete). Moreover, our results show that the positional encoding, a crucial component in ViTs, can be safely removed in our model. Experiments show that Vision Big Bird demonstrates competitive performance on common vision tasks.

Zhemin Zhang, Xun Gong

Recently, Transformers have shown promising performance in various vision tasks. However, the high costs of global self-attention remain challenging for Transformers, especially for high-resolution vision tasks. Local self-attention runs attention computation within a limited region for the sake of efficiency, resulting in insufficient context modeling as their receptive fields are small. In this work, we introduce two new attention modules to enhance the global modeling capability of the hierarchical vision transformer, namely, random sampling windows (RS-Win) and important region windows (IR-Win). Specifically, RS-Win sample random image patches to compose the window, following a uniform distribution, i.e., the patches in RS-Win can come from any position in the image. IR-Win composes the window according to the weights of the image patches in the attention map. Notably, RS-Win is able to capture global information throughout the entire model, even in earlier, high-resolution stages. IR-Win enables the self-attention module to focus on important regions of the image and capture more informative features. Incorporated with these designs, RSIR-Win Transformer demonstrates competitive performance on common vision tasks.

Zhemin Zhang, Xun Gong

Recently, Transformers have shown promising performance in various vision tasks. A challenging issue in Transformer design is that global self-attention is very expensive to compute, especially for the high-resolution vision tasks. Local self-attention performs attention computation within a local region to improve its efficiency, which leads to their receptive fields in a single attention layer are not large enough, resulting in insufficient context modeling. When observing a scene, humans usually focus on a local region while attending to non-attentional regions at coarse granularity. Based on this observation, we develop the axially expanded window self-attention mechanism that performs fine-grained self-attention within the local window and coarse-grained self-attention in the horizontal and vertical axes, and thus can effectively capturing both short- and long-range visual dependencies.

Jinyi Wu, Xun Gong, Zhemin Zhang

We propose Self-Supervised Implicit Attention (SSIA), a new approach that adaptively guides deep neural network models to gain attention by exploiting the properties of the models themselves. SSIA is a novel attention mechanism that does not require any extra parameters, computation, or memory access costs during inference, which is in contrast to existing attention mechanism. In short, by considering attention weights as higher-level semantic information, we reconsidered the implementation of existing attention mechanisms and further propose generating supervisory signals from higher network layers to guide lower network layers for parameter updates. We achieved this by building a self-supervised learning task using the hierarchical features of the network itself, which only works at the training stage. To verify the effectiveness of SSIA, we performed a particular implementation (called an SSIA block) in convolutional neural network models and validated it on several image classification datasets. The experimental results show that an SSIA block can significantly improve the model performance, even outperforms many popular attention methods that require additional parameters and computation costs, such as Squeeze-and-Excitation and Convolutional Block Attention Module. Our implementation will be available on GitHub.

Zhemin Zhang, Xun Gong

Treating class with a single center may hardly capture data distribution complexities. Using multiple sub-centers is an alternative way to address this problem. However, highly correlated sub-classes, the classifier's parameters grow linearly with the number of classes, and lack of intra-class compactness are three typical issues that need to be addressed in existing multi-subclass methods. To this end, we propose to use Fixed Sub-Center (F-SC), which allows the model to create more discrepant sub-centers while saving memory and cutting computational costs considerably. The F-SC specifically, first samples a class center Ui for each class from a uniform distribution, and then generates a normal distribution for each class, where the mean is equal to Ui. Finally, the sub-centers are sampled based on the normal distribution corresponding to each class, and the sub-centers are fixed during the training process avoiding the overhead of gradient calculation. Moreover, F-SC penalizes the Euclidean distance between the samples and their corresponding sub-centers, it helps remain intra-compactness. The experimental results show that F-SC significantly improves the accuracy of both image classification and fine-grained recognition tasks.

Qing Li, Huifang Feng, Xun Gong et al.

Estimating normals for noisy point clouds is a persistent challenge in 3D geometry processing, particularly for end-to-end oriented normal estimation. Existing methods generally address relatively clean data and rely on supervised priors to fit local surfaces within specific neighborhoods. In this paper, we propose a novel approach for learning normals from noisy point clouds through local gradient-aware surface filtering. Our method projects noisy points onto the underlying surface by utilizing normals and distances derived from an implicit function constrained by local gradients. We start by introducing a distance measurement operator for global surface fitting on noisy data, which integrates projected distances along normals. Following this, we develop an implicit field-based filtering approach for surface point construction, adding projection constraints on these points during filtering. To address issues of over-smoothing and gradient degradation, we further incorporate local gradient consistency constraints, as well as local gradient orientation and aggregation. Comprehensive experiments on normal estimation, surface reconstruction, and point cloud denoising demonstrate the state-of-the-art performance of our method. The source code and trained models are available at https://github.com/LeoQLi/LGSF.

Jingkun Yu, Guangkai Shang, Changtao Li et al.

Cancer pathological analysis requires modeling tumor heterogeneity across multiple modalities, primarily through transcriptomics and whole slide imaging (WSI), along with their spatial relations. On one hand, bulk transcriptomics and WSI images are largely available but lack spatial mapping; on the other hand, spatial transcriptomics (ST) data can offer high spatial resolution, yet facing challenges of high cost, low sequencing depth, and limited sample sizes. Therefore, the data foundation of either side is flawed and has its limit in accurately finding the mapping between the two modalities. To this end, we propose BiTro, a bidirectional transfer learning framework that can enhance bulk and spatial transcriptomics prediction from pathological images. Our contributions are twofold. First, we design a universal and transferable model architecture that works for both bulk+WSI and ST data. A major highlight is that we model WSI images on the cellular level to better capture cells' visual features, morphological phenotypes, and their spatial relations; to map cells' features to their transcriptomics measured in bulk or ST, we adopt multiple instance learning. Second, by using LoRA, our model can be efficiently transferred between bulk and ST data to exploit their complementary information. To test our framework, we conducted comprehensive experiments on five cancer datasets. Results demonstrate that 1) our base model can achieve better or competitive performance compared to existing models on bulk or spatial transcriptomics prediction, and 2) transfer learning can further improve the base model's performance.

Weiji Kong, Xun Gong, Juan Wang

Explaining the decisions of Deep Neural Networks (DNNs) for medical images has become increasingly important. Existing attribution methods have difficulty explaining the meaning of pixels while existing concept-based methods are limited by additional annotations or specific model structures that are difficult to apply to ultrasound images. In this paper, we propose the Lesion Concept Explainer (LCE) framework, which combines attribution methods with concept-based methods. We introduce the Segment Anything Model (SAM), fine-tuned on a large number of medical images, for concept discovery to enable a meaningful explanation of ultrasound image DNNs. The proposed framework is evaluated in terms of both faithfulness and understandability. We point out deficiencies in the popular faithfulness evaluation metrics and propose a new evaluation metric. Our evaluation of public and private breast ultrasound datasets (BUSI and FG-US-B) shows that LCE performs well compared to commonly-used explainability methods. Finally, we also validate that LCE can consistently provide reliable explanations for more meaningful fine-grained diagnostic tasks in breast ultrasound.

Jinrong Lv, Xun Gong, Zhaohuan Li et al.

Estimating Left Ventricular Ejection Fraction (LVEF) from echocardiograms constitutes an ill-posed inverse problem. Inherent noise, artifacts, and limited viewing angles introduce ambiguity, where a single video sequence may map not to a unique ground truth, but rather to a distribution of plausible physiological values. Prevailing deep learning approaches typically formulate this task as a standard regression problem that minimizes the Mean Squared Error (MSE). However, this paradigm compels the model to learn the conditional expectation, which may yield misleading predictions when the underlying posterior distribution is multimodal or heavy-tailed -- a common phenomenon in pathological scenarios. In this paper, we investigate the paradigm shift from deterministic regression toward generative regression. We propose the Multimodal Conditional Score-based Diffusion model for Regression (MCSDR), a probabilistic framework designed to model the continuous posterior distribution of LVEF conditioned on echocardiogram videos and patient demographic attribute priors. Extensive experiments conducted on the EchoNet-Dynamic, EchoNet-Pediatric, and CAMUS datasets demonstrate that MCSDR achieves state-of-the-art performance. Notably, qualitative analysis reveals that the generation trajectories of our model exhibit distinct behaviors in cases characterized by high noise or significant physiological variability, thereby offering a novel layer of interpretability for AI-aided diagnosis.

Jianhua Wu, Bingzhao Gao, Jincheng Gao et al.

With the development of artificial intelligence and breakthroughs in deep learning, large-scale Foundation Models (FMs), such as GPT, Sora, etc., have achieved remarkable results in many fields including natural language processing and computer vision. The application of FMs in autonomous driving holds considerable promise. For example, they can contribute to enhancing scene understanding and reasoning. By pre-training on rich linguistic and visual data, FMs can understand and interpret various elements in a driving scene, and provide cognitive reasoning to give linguistic and action instructions for driving decisions and planning. Furthermore, FMs can augment data based on the understanding of driving scenarios to provide feasible scenes of those rare occurrences in the long tail distribution that are unlikely to be encountered during routine driving and data collection. The enhancement can subsequently lead to improvement in the accuracy and reliability of autonomous driving systems. Another testament to the potential of FMs' applications lies in World Models, exemplified by the DREAMER series, which showcases the ability to comprehend physical laws and dynamics. Learning from massive data under the paradigm of self-supervised learning, World Model can generate unseen yet plausible driving environments, facilitating the enhancement in the prediction of road users' behaviors and the off-line training of driving strategies. In this paper, we synthesize the applications and future trends of FMs in autonomous driving. By utilizing the powerful capabilities of FMs, we strive to tackle the potential issues stemming from the long-tail distribution in autonomous driving, consequently advancing overall safety in this domain.

Jiawei Liu, Xun Gong, Fen Fang et al.

Most Human-Machine Interaction (HMI) research overlooks the maneuvering needs of passengers in autonomous driving (AD). Natural language offers an intuitive interface, yet translating passenger open-ended instructions into control signals, without sacrificing interpretability and traceability, remains a challenge. This study proposes an instruction-realization framework that leverages a large language model (LLM) to interpret instructions, generates executable scripts that schedule multiple model predictive control (MPC)-based motion planners based on real-time feedback, and converts planned trajectories into control signals. This scheduling-centric design decouples semantic reasoning from vehicle control at different timescales, establishing a transparent, traceable decision-making chain from high-level instructions to low-level actions. Due to the absence of high-fidelity evaluation tools, this study introduces a benchmark for open-ended instruction realization in a closed-loop setting. Comprehensive experiments reveal that the framework significantly improves task-completion rates over instruction-realization baselines, reduces LLM query costs, achieves safety and compliance on par with specialized AD approaches, and exhibits considerable tolerance to LLM inference latency. For more qualitative illustrations and a clearer understanding.

Xiao Li, Xiang Zheng, Yifeng Gao et al.

Embodied Artificial Intelligence (Embodied AI) integrates perception, cognition, planning, and interaction into agents that operate in open-world, safety-critical environments. As these systems gain autonomy and enter domains such as transportation, healthcare, and industrial or assistive robotics, ensuring their safety becomes both technically challenging and socially indispensable. Unlike digital AI systems, embodied agents must act under uncertain sensing, incomplete knowledge, and dynamic human-robot interactions, where failures can directly lead to physical harm. This survey provides a comprehensive and structured review of safety research in embodied AI, examining attacks and defenses across the full embodied pipeline, from perception and cognition to planning, action and interaction, and agentic system. We introduce a multi-level taxonomy that unifies fragmented lines of work and connects embodied-specific safety findings with broader advances in vision, language, and multimodal foundation models. Our review synthesizes insights from over 400 papers spanning adversarial, backdoor, jailbreak, and hardware-level attacks; attack detection, safe training and robust inference; and risk-aware human-agent interaction. This analysis reveals several overlooked challenges, including the fragility of multimodal perception fusion, the instability of planning under jailbreak attacks, and the trustworthiness of human-agent interaction in open-ended scenarios. By organizing the field into a coherent framework and identifying critical research gaps, this survey provides a roadmap for building embodied agents that are not only capable and autonomous but also safe, robust, and reliable in real-world deployment.

Wei Li, Xun Gong, Jiao Li et al.

Learning medical visual representations from paired images and reports is a promising direction in representation learning. However, current vision-language pretraining methods in the medical domain often simplify clinical reports into single entities or fragmented tokens, ignoring their inherent structure. In addition, contrastive learning frameworks typically depend on large quantities of hard negative samples, which is impractical for small-scale medical datasets. To tackle these challenges, we propose Adaptive Grouped Alignment (AGA), a new framework that captures structured semantics from paired medical images and reports. AGA introduces a bidirectional grouping mechanism based on a sparse similarity matrix. For each image-report pair, we compute fine-grained similarities between text tokens and image patches. Each token selects its top-matching patches to form a visual group, and each patch selects its most related tokens to form a language group. To enable adaptive grouping, we design two threshold gating modules, called Language Grouped Threshold Gate and Vision Grouped Threshold Gate, which learn grouping thresholds dynamically. Group representations are computed as weighted averages based on similarity scores. To align each token with its group representation, we introduce an Instance Aware Group Alignment loss that operates within each image-text pair, removing the need for external negatives. Finally, a Bidirectional Cross-modal Grouped Alignment module is applied to enhance fine-grained alignment between visual and linguistic group representations. Extensive experiments on public and private datasets show that our method achieves strong performance on image-text retrieval and classification tasks under both fine-tuning and zero-shot settings.

Xinyu Xia, Xingjun Ma, Yunfeng Hu et al.

Ensuring robust and generalizable autonomous driving requires not only broad scenario coverage but also efficient repair of failure cases, particularly those related to challenging and safety-critical scenarios. However, existing scenario generation and selection methods often lack adaptivity and semantic relevance, limiting their impact on performance improvement. In this paper, we propose \textbf{SERA}, an LLM-powered framework that enables autonomous driving systems to self-evolve by repairing failure cases through targeted scenario recommendation. By analyzing performance logs, SERA identifies failure patterns and dynamically retrieves semantically aligned scenarios from a structured bank. An LLM-based reflection mechanism further refines these recommendations to maximize relevance and diversity. The selected scenarios are used for few-shot fine-tuning, enabling targeted adaptation with minimal data. Experiments on the benchmark show that SERA consistently improves key metrics across multiple autonomous driving baselines, demonstrating its effectiveness and generalizability under safety-critical conditions.

Jie Zhang, Xun Gong, Zhonglin Sun

Face recognition has made remarkable strides, driven by the expanding scale of datasets, advancements in various backbone and discriminative losses. However, face recognition performance is heavily affected by the label noise, especially closed-set noise. While numerous studies have focused on handling label noise, addressing closed-set noise still poses challenges. This paper identifies this challenge as training isn't robust to noise at the early-stage training, and necessitating an appropriate learning strategy for samples with low confidence, which are often misclassified as closed-set noise in later training phases. To address these issues, we propose a new framework to stabilize the training at early stages and split the samples into clean, ambiguous and noisy groups which are devised with separate training strategies. Initially, we employ generated auxiliary closed-set noisy samples to enable the model to identify noisy data at the early stages of training. Subsequently, we introduce how samples are split into clean, ambiguous and noisy groups by their similarity to the positive and nearest negative centers. Then we perform label fusion for ambiguous samples by incorporating accumulated model predictions. Finally, we apply label smoothing within the closed set, adjusting the label to a point between the nearest negative class and the initially assigned label. Extensive experiments validate the effectiveness of our method on mainstream face datasets, achieving state-of-the-art results. The code will be released upon acceptance.

Lin Fan, Xun Gong, Cenyang Zheng et al.

The intersection of medical Visual Question Answering (Med-VQA) is a challenging research topic with advantages including patient engagement and clinical expert involvement for second opinions. However, existing Med-VQA methods based on joint embedding fail to explain whether their provided results are based on correct reasoning or coincidental answers, which undermines the credibility of VQA answers. In this paper, we investigate the construction of a more cohesive and stable Med-VQA structure. Motivated by causal effect, we propose a novel Triangular Reasoning VQA (Tri-VQA) framework, which constructs reverse causal questions from the perspective of "Why this answer?" to elucidate the source of the answer and stimulate more reasonable forward reasoning processes. We evaluate our method on the Endoscopic Ultrasound (EUS) multi-attribute annotated dataset from five centers, and test it on medical VQA datasets. Experimental results demonstrate the superiority of our approach over existing methods. Our codes and pre-trained models are available at https://anonymous.4open.science/r/Tri_VQA.

Lin Fan, Yafei Ou, Cenyang Zheng et al.

Multi-modal learning has shown exceptional performance in various tasks, especially in medical applications, where it integrates diverse medical information for comprehensive diagnostic evidence. However, there still are several challenges in multi-modal learning, 1. Heterogeneity between modalities, 2. uncertainty in missing modalities, 3. influence of intrinsic noise, and 4. interpretability for fusion result. This paper introduces the Modal-Domain Attention (MDA) model to address the above challenges. MDA constructs linear relationships between modalities through continuous attention, due to its ability to adaptively allocate dynamic attention to different modalities, MDA can reduce attention to low-correlation data, missing modalities, or modalities with inherent noise, thereby maintaining SOTA performance across various tasks on multiple public datasets. Furthermore, our observations on the contribution of different modalities indicate that MDA aligns with established clinical diagnostic imaging gold standards and holds promise as a reference for pathologies where these standards are not yet clearly defined. The code and dataset will be available.

Hang Shao, Bei Liu, Wei Wang et al.

As a popular multilingual and multitask pre-trained speech model, Whisper has the problem of curse of multilinguality. To enhance multilingual capabilities in small Whisper models, we propose DQ-Whisper, a novel joint distillation and quantization framework to compress Whisper for efficient inference. Firstly, we propose a novel dynamic matching distillation strategy. Then, a quantization-aware distillation framework is introduced to integrate quantization with distillation. Experimental results on various multilingual datasets show that our suggested distillation approach can effectively enhance the multilingual capabilities of small Whisper models without increasing computational costs. Up to 5.18x reduction in model size is achieved with marginal performance degradation. In addition, quantization is compatible with distillation, which can result in a higher compression rate.

Zhemin Zhang, Xun Gong, Jinyi Wu

Attention mechanism, being frequently used to train networks for better feature representations, can effectively disentangle the target object from irrelevant objects in the background. Given an arbitrary image, we find that the background's irrelevant objects are most likely to occlude/block the target object. We propose, based on this finding, a ReplaceBlock to simulate the situations when the target object is partially occluded by the objects that are deemed as background. Specifically, ReplaceBlock erases the target object in the image, and then generates a feature map with only irrelevant objects and background by the model. Finally, some regions in the background feature map are used to replace some regions of the target object in the original image feature map. In this way, ReplaceBlock can effectively simulate the feature map of the occluded image. The experimental results show that ReplaceBlock works better than DropBlock in regularizing convolutional networks.

Shengdong Du, Tianrui Li, Xun Gong et al.

Traffic flow forecasting has been regarded as a key problem of intelligent transport systems. In this work, we propose a hybrid multimodal deep learning method for short-term traffic flow forecasting, which can jointly and adaptively learn the spatial-temporal correlation features and long temporal interdependence of multi-modality traffic data by an attention auxiliary multimodal deep learning architecture. According to the highly nonlinear characteristics of multi-modality traffic data, the base module of our method consists of one-dimensional Convolutional Neural Networks (1D CNN) and Gated Recurrent Units (GRU) with the attention mechanism. The former is to capture the local trend features and the latter is to capture the long temporal dependencies. Then, we design a hybrid multimodal deep learning framework (HMDLF) for fusing share representation features of different modality traffic data by multiple CNN-GRU-Attention modules. The experimental results indicate that the proposed multimodal deep learning model is capable of dealing with complex nonlinear urban traffic flow forecasting with satisfying accuracy and effectiveness.

Zeng Yu, Tianrui Li, Ning Yu et al.

This paper aims to develop a new architecture that can make full use of the feature maps of convolutional networks. To this end, we study a number of methods for video-based person re-identification and make the following findings: 1) Max-pooling only focuses on the maximum value of a receptive field, wasting a lot of information. 2) Networks with different streams even including the one with the worst performance work better than networks with same streams, where each one has the best performance alone. 3) A full connection layer at the end of convolutional networks is not necessary. Based on these studies, we propose a new convolutional architecture termed Three-Stream Convolutional Networks (TSCN). It first uses different streams to learn different aspects of feature maps for attentive spatio-temporal fusion of video, and then merges them together to study some union features. To further utilize the feature maps, two architectures are designed by using the strategies of multi-scale and upsampling. Comparative experiments on iLIDS-VID, PRID-2011 and MARS datasets illustrate that the proposed architectures are significantly better for feature extraction than the state-of-the-art models.

Xun Gong, Negar Kiyavash, Parv Venkitasubramaniam

Timing side channels in two-user schedulers are studied. When two users share a scheduler, one user may learn the other user's behavior from patterns of service timings. We measure the information leakage of the resulting timing side channel in schedulers serving a legitimate user and a malicious attacker, using a privacy metric defined as the Shannon equivocation of the user's job density. We show that the commonly used first-come-first-serve (FCFS) scheduler provides no privacy as the attacker is able to to learn the user's job pattern completely. Furthermore, we introduce an scheduling policy, accumulate-and-serve scheduler, which services jobs from the user and attacker in batches after buffering them. The information leakage in this scheduler is mitigated at the price of service delays, and the maximum privacy is achievable when large delays are added.

Xun Gong, Mavis Rodrigues, Negar Kiyavash

Flow watermarks efficiently link packet flows in a network in order to thwart various attacks such as stepping stones. We study the problem of designing good flow watermarks. Earlier flow watermarking schemes mostly considered substitution errors, neglecting the effects of packet insertions and deletions that commonly happen within a network. More recent schemes consider packet deletions but often at the expense of the watermark visibility. We present an invisible flow watermarking scheme capable of enduring a large number of packet losses and insertions. To maintain invisibility, our scheme uses quantization index modulation (QIM) to embed the watermark into inter-packet delays, as opposed to time intervals including many packets. As the watermark is injected within individual packets, packet losses and insertions may lead to watermark desynchronization and substitution errors. To address this issue, we add a layer of error-correction coding to our scheme. Experimental results on both synthetic and real network traces demonstrate that our scheme is robust to network jitter, packet drops and splits, while remaining invisible to an attacker.

Qiyan Wang, Xun Gong, Giang T. K. Nguyen et al.

A key challenge in censorship-resistant web browsing is being able to direct legitimate users to redirection proxies while preventing censors, posing as insiders, from discovering their addresses and blocking them. We propose a new framework for censorship-resistant web browsing called {\it CensorSpoofer} that addresses this challenge by exploiting the asymmetric nature of web browsing traffic and making use of IP spoofing. CensorSpoofer de-couples the upstream and downstream channels, using a low-bandwidth indirect channel for delivering outbound requests (URLs) and a high-bandwidth direct channel for downloading web content. The upstream channel hides the request contents using steganographic encoding within email or instant messages, whereas the downstream channel uses IP address spoofing so that the real address of the proxies is not revealed either to legitimate users or censors. We built a proof-of-concept prototype that uses encrypted VoIP for this downstream channel and demonstrated the feasibility of using the CensorSpoofer framework in a realistic environment.