Peng Cao

Kaiwen Zha, Peng Cao, Jeany Son et al.

Deep regression models typically learn in an end-to-end fashion without explicitly emphasizing a regression-aware representation. Consequently, the learned representations exhibit fragmentation and fail to capture the continuous nature of sample orders, inducing suboptimal results across a wide range of regression tasks. To fill the gap, we propose Rank-N-Contrast (RNC), a framework that learns continuous representations for regression by contrasting samples against each other based on their rankings in the target space. We demonstrate, theoretically and empirically, that RNC guarantees the desired order of learned representations in accordance with the target orders, enjoying not only better performance but also significantly improved robustness, efficiency, and generalization. Extensive experiments using five real-world regression datasets that span computer vision, human-computer interaction, and healthcare verify that RNC achieves state-of-the-art performance, highlighting its intriguing properties including better data efficiency, robustness to spurious targets and data corruptions, and generalization to distribution shifts. Code is available at: https://github.com/kaiwenzha/Rank-N-Contrast.

Zhiqiang Shen, Peng Cao, Hua Yang et al.

Consistency regularization and pseudo labeling-based semi-supervised methods perform co-training using the pseudo labels from multi-view inputs. However, such co-training models tend to converge early to a consensus, degenerating to the self-training ones, and produce low-confidence pseudo labels from the perturbed inputs during training. To address these issues, we propose an Uncertainty-guided Collaborative Mean-Teacher (UCMT) for semi-supervised semantic segmentation with the high-confidence pseudo labels. Concretely, UCMT consists of two main components: 1) collaborative mean-teacher (CMT) for encouraging model disagreement and performing co-training between the sub-networks, and 2) uncertainty-guided region mix (UMIX) for manipulating the input images according to the uncertainty maps of CMT and facilitating CMT to produce high-confidence pseudo labels. Combining the strengths of UMIX with CMT, UCMT can retain model disagreement and enhance the quality of pseudo labels for the co-training segmentation. Extensive experiments on four public medical image datasets including 2D and 3D modalities demonstrate the superiority of UCMT over the state-of-the-art. Code is available at: https://github.com/Senyh/UCMT.

Junming Su, Zhiqiang Shen, Peng Cao et al.

The existing barely-supervised medical image segmentation (BSS) methods, adopting a registration-segmentation paradigm, aim to learn from data with very few annotations to mitigate the extreme label scarcity problem. However, this paradigm poses a challenge: pseudo-labels generated by image registration come with significant noise. To address this issue, we propose a self-paced sample selection framework (SPSS) for BSS. Specifically, SPSS comprises two main components: 1) self-paced uncertainty sample selection (SU) for explicitly improving the quality of pseudo labels in the image space, and 2) self-paced bidirectional feature contrastive learning (SC) for implicitly improving the quality of pseudo labels through enhancing the separability between class semantics in the feature space. Both SU and SC are trained collaboratively in a self-paced learning manner, ensuring that SPSS can learn from high-quality pseudo labels for BSS. Extensive experiments on two public medical image segmentation datasets demonstrate the effectiveness and superiority of SPSS over the state-of-the-art. Our code is release at https://github.com/SuuuJM/SPSS.

Hao He, Yuan Yuan, Ying-Cong Chen et al.

With the increasing popularity of telehealth, it becomes critical to ensure that basic physiological signals can be monitored accurately at home, with minimal patient overhead. In this paper, we propose a contactless approach for monitoring patients' blood oxygen at home, simply by analyzing the radio signals in the room, without any wearable devices. We extract the patients' respiration from the radio signals that bounce off their bodies and devise a novel neural network that infers a patient's oxygen estimates from their breathing signal. Our model, called \emph{Gated BERT-UNet}, is designed to adapt to the patient's medical indices (e.g., gender, sleep stages). It has multiple predictive heads and selects the most suitable head via a gate controlled by the person's physiological indices. Extensive empirical results show that our model achieves high accuracy on both medical and radio datasets.

Siyu Liu, Guangqi Wen, Peng Cao et al.

Modeling the complex interactions among functional subnetworks is crucial for the diagnosis of mental disorders and the identification of functional pathways. However, learning the interactions of the underlying subnetworks remains a significant challenge for existing Transformer-based methods due to the limited number of training samples. To address these challenges, we propose KD-Brain, a Prior-Informed Graph Learning framework for explicitly encoding prior knowledge to guide the learning process. Specifically, we design a Semantic-Conditioned Interaction mechanism that injects semantic priors into the attention query, explicitly navigating the subnetwork interactions based on their functional identities. Furthermore, we introduce a Pathology-Consistent Constraint, which regularizes the model optimization by aligning the learned interaction distributions with clinical priors. Additionally, KD-Brain leads to state-of-the-art performance on a wide range of disorder diagnosis tasks and identifies interpretable biomarkers consistent with psychiatric pathophysiology. Our code is available at https://anonymous.4open.science/r/KDBrain.

Xiaolong Li, Guiliang Guo, Guangqi Wen et al.

Mental disorder populations exhibit pronounced heterogeneity -- that is, the significant differences between samples -- poses a significant challenge to the definition of positive pairs in contrastive learning. To address this, we propose a subtype-guided contrastive learning framework that models patient heterogeneity as latent subtypes and incorporates them as structural priors to guide discriminative representation learning. Specifically, we construct multi-view representations by combining patients' clinical text with graph structure adaptively learned from BOLD signals, to uncover latent subtypes via unsupervised spectral clustering. A dual-level attention mechanism is proposed to construct prototypes for capturing stable subtype-specific connectivity patterns. We further propose a subtype-guided contrastive learning strategy that pulls samples toward their subtype prototype graph, reinforcing intra-subtype consistency for providing effective supervisory signals to improve model performance. We evaluate our method on Major Depressive Disorder (MDD), Bipolar Disorder (BD), and Autism Spectrum Disorders (ASD). Experimental results confirm the effectiveness of subtype prototype graphs in guiding contrastive learning and demonstrate that the proposed approach outperforms state-of-the-art approaches. Our code is available at https://anonymous.4open.science/r/BrainSCL-06D7.

Qingshan Hou, Peng Cao, Jiaqi Wang et al.

Retinal fundus images have been applied for the diagnosis and screening of eye diseases, such as Diabetic Retinopathy (DR) or Diabetic Macular Edema (DME). However, both low-quality fundus images and style inconsistency potentially increase uncertainty in the diagnosis of fundus disease and even lead to misdiagnosis by ophthalmologists. Most of the existing image enhancement methods mainly focus on improving the image quality by leveraging the guidance of high-quality images, which is difficult to be collected in medical applications. In this paper, we tackle image quality enhancement in a fully unsupervised setting, i.e., neither paired images nor high-quality images. To this end, we explore the potential of the self-supervised task for improving the quality of fundus images without the requirement of high-quality reference images. Specifically, we construct multiple patch-wise domains via an auxiliary pre-trained quality assessment network and a style clustering. To achieve robust low-quality image enhancement and address style inconsistency, we formulate two self-supervised domain adaptation tasks to disentangle the features of image content, low-quality factor and style information by exploring intrinsic supervision signals within the low-quality images. Extensive experiments are conducted on EyeQ and Messidor datasets, and results show that our DASQE method achieves new state-of-the-art performance when only low-quality images are available.

Guanxiong Luo, Mengmeng Kuang, Peng Cao · tencent-ai

The data consistency for the physical forward model is crucial in inverse problems, especially in MR imaging reconstruction. The standard way is to unroll an iterative algorithm into a neural network with a forward model embedded. The forward model always changes in clinical practice, so the learning component's entanglement with the forward model makes the reconstruction hard to generalize. The deep learning-based proximal gradient descent was proposed and use a network as regularization term that is independent of the forward model, which makes it more generalizable for different MR acquisition settings. This one-time pre-trained regularization is applied to different MR acquisition settings and was compared to conventional L1 regularization showing ~3 dB improvement in the peak signal-to-noise ratio. We also demonstrated the flexibility of the proposed method in choosing different undersampling patterns.

Zhiqiang Shen, Peng Cao, Junming Su et al.

Mix-up is a key technique for consistency regularization-based semi-supervised learning methods, blending two or more images to generate strong-perturbed samples for strong-weak pseudo supervision. Existing mix-up operations are performed either randomly or with predefined fixed rules, such as replacing low-confidence patches with high-confidence ones. The former lacks control over the perturbation degree, leading to overfitting on randomly perturbed samples, while the latter tends to generate images with trivial perturbations, both of which limit the effectiveness of consistency regularization. This paper aims to answer the following question: How can image mix-up perturbation be adaptively performed during training? To this end, we propose an Adaptive Mix algorithm (AdaMix) for image mix-up in a self-paced learning manner. Given that, in general, a model's performance gradually improves during training, AdaMix is equipped with a self-paced curriculum that, in the initial training stage, provides relatively simple perturbed samples and then gradually increases the difficulty of perturbed images by adaptively controlling the perturbation degree based on the model's learning state estimated by a self-paced regularize. We develop three frameworks with our AdaMix, i.e., AdaMix-ST, AdaMix-MT, and AdaMix-CT, for semi-supervised medical image segmentation. Extensive experiments on three public datasets show that the proposed frameworks can achieve superior performance. For example, compared with the state-of-the-art, AdaMix-CT achieves relative improvements of 2.62% in Dice similarity coefficient and 48.25% in average surface distance on the ACDC dataset with 10% labeled data. The results demonstrate that mix-up operations with dynamically adjusted perturbation strength based on the segmentation model's state can significantly enhance the effectiveness of consistency regularization.

Tianle Zhang, Zhihao Yuan, Dafeng Chi et al.

Robotic autonomy in open-world environments is fundamentally limited by insufficient data diversity and poor cross-embodiment generalization. Existing robotic datasets are often limited in scale and task coverage, while relatively large differences across robot embodiments impede effective behavior knowledge transfer. To address these challenges, we propose JoyAI-RA, a vision-language-action (VLA) embodied foundation model tailored for generalizable robotic manipulation. JoyAI-RA presents a multi-source multi-level pretraining framework that integrates web data, large-scale egocentric human manipulation videos, simulation-generated trajectories, and real-robot data. Through training on heterogeneous multi-source data with explicit action-space unification, JoyAI-RA effectively bridges embodiment gaps, particularly between human manipulation and robotic control, thereby enhancing cross-embodiment behavior learning. JoyAI-RA outperforms state-of-the-art methods in both simulation and real-world benchmarks, especially on diverse tasks with generalization demands.

Xin Huang, Junjie Liang, Qingshan Hou et al.

Medical image synthesis is crucial for alleviating data scarcity and privacy constraints. However, fine-tuning general text-to-image (T2I) models remains challenging, mainly due to the significant modality gap between complex visual details and abstract clinical text. In addition, semantic entanglement persists, where coarse-grained text embeddings blur the boundary between anatomical structures and imaging styles, thus weakening controllability during generation. To address this, we propose a Visually-Guided Text Disentanglement framework. We introduce a cross-modal latent alignment mechanism that leverages visual priors to explicitly disentangle unstructured text into independent semantic representations. Subsequently, a Hybrid Feature Fusion Module (HFFM) injects these features into a Diffusion Transformer (DiT) through separated channels, enabling fine-grained structural control. Experimental results in three datasets demonstrate that our method outperforms existing approaches in terms of generation quality and significantly improves performance on downstream classification tasks. The source code is available at https://github.com/hx111/VG-MedGen.

Shunlai Cui, Peng Cao, Yuan Zhu et al.

LiDAR model selection is a critical issue in roadside sensing systems, as it directly determines both perception capability and deployment cost. However, the lack of empirical benchmarks for comparing perception performance across different LiDAR configurations has greatly constrained scientific sensor selection and deployment planning. To address this gap, we present MR-LiDAR, a controlled multi-resolution LiDAR benchmark for roadside perception diagnostics. Using 16-, 32-, 80-, and 128-beam LiDARs in identical roadside scenarios, we collect point clouds and ground-truth annotations for diverse traffic participants, including vehicles and vulnerable road users (VRUs), across varying distances. This controlled design isolates intrinsic LiDAR specifications, particularly beam count and beam distribution, as the key variables for precise performance diagnostics. Based on MR-LiDAR, we conduct systematic empirical analyses to examine how beam count, beam distribution, target distance, object category, and vehicle occlusion affect LiDAR perception performance. The results reveal that all of these factors have substantial impacts. In particular, contrary to the common assumption that higher beam counts always yield better perception, we show that an 80-beam LiDAR with optimized beam distribution can match or even outperform a 128-beam LiDAR with uniform beam distribution. In addition, we provide a practical reference guide for LiDAR selection, including target point-count statistics and detection performance comparisons based on two widely used detection algorithms. This work offers a diagnostic benchmark and practical guidance for determining cost-effective LiDAR configurations in roadside perception applications.

Chiyu Zhang, Jun Yang, Zaiyan Dai et al.

In recent years, arbitrary image style transfer has attracted more and more attention. Given a pair of content and style images, a stylized one is hoped that retains the content from the former while catching style patterns from the latter. However, it is difficult to simultaneously keep well the trade-off between the content details and the style features. To stylize the image with sufficient style patterns, the content details may be damaged and sometimes the objects of images can not be distinguished clearly. For this reason, we present a new transformer-based method named STT for image style transfer and an edge loss which can enhance the content details apparently to avoid generating blurred results for excessive rendering on style features. Qualitative and quantitative experiments demonstrate that STT achieves comparable performance to state-of-the-art image style transfer methods while alleviating the content leak problem.

Guiliang Guo, Guangqi Wen, Lingwen Liu et al.

Dynamic functional connectivity captures time-varying brain states for better neuropsychiatric diagnosis and spatio-temporal interpretability, i.e., identifying when discriminative disease signatures emerge and where they reside in the connectivity topology. Reliable interpretability faces major challenges: diagnostic signals are often subtle and sparsely distributed across both time and topology, while nuisance fluctuations and non-diagnostic connectivities are pervasive. To address these issues, we propose BrainSTR, a spatio-temporal contrastive learning framework for interpretable dynamic brain network modeling. BrainSTR learns state-consistent phase boundaries via a data-driven Adaptive Phase Partition module, identifies diagnostically critical phases with attention, and extracts disease-related connectivity within each phase using an Incremental Graph Structure Generator regularized by binarization, temporal smoothness, and sparsity. Then, we introduce a spatio-temporal supervised contrastive learning approach that leverages diagnosis-relevant spatio-temporal patterns to refine the similarity metric between samples and capture more discriminative spatio-temporal features, thereby constructing a well-structured semantic space for coherent and interpretable representations. Experiments on ASD, BD, and MDD validate the effectiveness of BrainSTR, and the discovered critical phases and subnetworks provide interpretable evidence consistent with prior neuroimaging findings. Our code: https://anonymous.4open.science/r/BrainSTR1.

Jingfeng Tang, Peng Cao, Guangqi Wen et al.

Brain network analysis based on functional Magnetic Resonance Imaging (fMRI) is pivotal for diagnosing brain disorders. Existing approaches typically rely on predefined functional sub-networks to construct sub-network associations. However, we identified many cross-network interaction patterns with high Pearson correlations that this strict, prior-based organization fails to capture. To overcome this limitation, we propose the Brain Hierarchical Organization Learning (BrainHO) to learn inherently hierarchical brain network dependencies based on their intrinsic features rather than predefined sub-network labels. Specifically, we design a hierarchical attention mechanism that allows the model to aggregate nodes into a hierarchical organization, effectively capturing intricate connectivity patterns at the subgraph level. To ensure diverse, complementary, and stable organizations, we incorporate an orthogonality constraint loss, alongside a hierarchical consistency constraint strategy, to refine node-level features using high-level graph semantics. Extensive experiments on the publicly available ABIDE and REST-meta-MDD datasets demonstrate that BrainHO not only achieves state-of-the-art classification performance but also uncovers interpretable, clinically significant biomarkers by precisely localizing disease-related sub-networks.

Yihang Li, Xuelong Wei, Jingzhou Luo et al.

The advancement of robot learning is currently hindered by the scarcity of large-scale, high-quality datasets. While established data collection methods such as teleoperation and universal manipulation interfaces dominate current datasets, they suffer from inherent limitations in scalability and real-world deployability. Human egocentric video collection, by contrast, has emerged as a promising approach to enable scalable, natural and in-the-wild data collection. As such, we present EgoLive, a large-scale, high-quality egocentric dataset designed explicitly for robot manipulation learning. EgoLive establishes three distinctive technical advantages over existing egocentric datasets: first, it represents the largest open-source annotated egocentric dataset focused on real-world task-oriented human routines to date; second, it delivers leading data quality via a customized head-mounted capture device and comprehensive high-precision multi-modal annotations; third, all data is collected exclusively in unconstrained real-world scenarios and encompasses vertical field human working data, including home service, retail, and other practical work scenarios, providing superior diversity and ecological validity. With the introduction of EgoLive, we aim to provide the research community with a scalable, high-quality dataset that accelerates breakthroughs in generalizable robotic models and facilitates the real-world deployment of robot systems.

Chongxiao Wang, Junjie Liang, Peng Cao et al.

Depression is a severe mental disorder, and reliable identification plays a critical role in early intervention and treatment. Multimodal depression detection aims to improve diagnostic performance by jointly modeling complementary information from multiple modalities. Recently, numerous multimodal learning approaches have been proposed for depression analysis; however, these methods suffer from the following limitations: 1) inter-modal inconsistency and depression-unrelated interference, where depression-related cues may conflict across modalities while substantial irrelevant content obscures critical depressive signals, and 2) diverse individual depressive presentations, leading to individual differences in modality and cue importance that hinder reliable fusion. To address these issues, we propose Individual-aware Multimodal Depression-related Representation Learning Framework (IDRL) for robust depression diagnosis. Specifically, IDRL 1) disentangles multimodal representations into a modality-common depression space, a modality-specific depression space, and a depression-unrelated space to enhance modality alignment while suppressing irrelevant information, and 2) introduces an individual-aware modality-fusion module (IAF) that dynamically adjusts the weights of disentangled depression-related features based on their predictive significance, thereby achieving adaptive cross-modal fusion for different individuals. Extensive experiments demonstrate that IDRL achieves superior and robust performance for multimodal depression detection.

Chunzhen Jin, Yongfeng Huang, Yaqi Wang et al.

Text style transfer, an important research direction in natural language processing, aims to adapt the text to various preferences but often faces challenges with limited resources. In this work, we introduce a novel method termed Style Extraction and Tunable Inference via Dual-level Transferable Prompt Learning (SETTP) for effective style transfer in low-resource scenarios. First, SETTP learns source style-level prompts containing fundamental style characteristics from high-resource style transfer. During training, the source style-level prompts are transferred through an attention module to derive a target style-level prompt for beneficial knowledge provision in low-resource style transfer. Additionally, we propose instance-level prompts obtained by clustering the target resources based on the semantic content to reduce semantic bias. We also propose an automated evaluation approach of style similarity based on alignment with human evaluations using ChatGPT-4. Our experiments across three resourceful styles show that SETTP requires only 1/20th of the data volume to achieve performance comparable to state-of-the-art methods. In tasks involving scarce data like writing style and role style, SETTP outperforms previous methods by 16.24\%.

Xi Chen, Zhiqiang Shen, Peng Cao et al.

Medical images are usually collected from multiple domains, leading to domain shifts that impair the performance of medical image segmentation models. Domain Generalization (DG) aims to address this issue by training a robust model with strong generalizability. Recently, numerous domain randomization-based DG methods have been proposed. However, these methods suffer from the following limitations: 1) constrained efficiency of domain randomization due to their exclusive dependence on image style perturbation, and 2) neglect of the adverse effects of over-augmented images on model training. To address these issues, we propose a novel domain randomization-based DG method, called content style augmentation (ConStyX), for generalizable medical image segmentation. Specifically, ConStyX 1) augments the content and style of training data, allowing the augmented training data to better cover a wider range of data domains, and 2) leverages well-augmented features while mitigating the negative effects of over-augmented features during model training. Extensive experiments across multiple domains demonstrate that our ConStyX achieves superior generalization performance. The code is available at https://github.com/jwxsp1/ConStyX.

Jianqiang Lin, Zhiqiang Shen, Peng Cao et al.

Although diffusion models have achieved remarkable progress in multi-modal magnetic resonance imaging (MRI) translation tasks, existing methods still tend to suffer from anatomical inconsistencies or degraded texture details when handling arbitrary missing-modality scenarios. To address these issues, we propose a latent diffusion-based multi-modal MRI translation framework, termed MSG-LDM. By leveraging the available modalities, the proposed method infers complete structural information, which preserves reliable boundary details. Specifically, we introduce a style--structure disentanglement mechanism in the latent space, which explicitly separates modality-specific style features from shared structural representations, and jointly models low-frequency anatomical layouts and high-frequency boundary details in a multi-scale feature space. During the structure disentanglement stage, high-frequency structural information is explicitly incorporated to enhance feature representations, guiding the model to focus on fine-grained structural cues while learning modality-invariant low-frequency anatomical representations. Furthermore, to reduce interference from modality-specific styles and improve the stability of structure representations, we design a style consistency loss and a structure-aware loss. Extensive experiments on the BraTS2020 and WMH datasets demonstrate that the proposed method outperforms existing MRI synthesis approaches, particularly in reconstructing complete structures. The source code is publicly available at https://github.com/ziyi-start/MSG-LDM.

Peng Cao, Ali Mirzazadeh, Jong Woo Lee et al.

Clinical EEG interpretation requires reasoning over full EEG sessions and integrating signal patterns with clinical context. Existing EEG foundation models are largely designed for short-window decoding and do not incorporate clinical context. We introduce CLEF, a clinically grounded long-context EEG foundation model. CLEF represents EEG sessions as 3D multitaper spectrogram tokens, enabling tractable Transformer modeling at session scale, and aligns embeddings with neurologist reports and structured EHR data through contrastive objectives. We evaluate CLEF on a new 234-task benchmark spanning disease phenotypes, medication exposures, and EEG findings, with more than 260k EEG sessions from over 108k patients. CLEF outperforms prior EEG foundation models on 229 of 234 tasks, improving mean AUROC from 0.65 to 0.74. Reconstruction-only pretraining surpasses prior EEG foundation models, while report and EHR alignment yields further gains. Held-out concept and external-cohort experiments suggest that these representations transfer beyond observed alignment targets. These results support session-scale, clinically grounded representation learning as a promising foundation-model paradigm for clinical EEG.

Jiarun Liu, Shiyue Xu, Shangkun Liu et al.

Large Language Models are increasingly deployed as autonomous agents for complex real-world tasks, yet existing systems often focus on isolated improvements without a unifying design for robustness and adaptability. We propose a generalist agent architecture that integrates three core components: a collective multi-agent framework combining planning and execution agents with critic model voting, a hierarchical memory system spanning working, semantic, and procedural layers, and a refined tool suite for search, code execution, and multimodal parsing. Evaluated on a comprehensive benchmark, our framework consistently outperforms open-source baselines and approaches the performance of proprietary systems. These results demonstrate the importance of system-level integration and highlight a path toward scalable, resilient, and adaptive AI assistants capable of operating across diverse domains and tasks.

Zhiqiang Shen, Peng Cao, Xiaoli Liu et al.

Label scarcity remains a major challenge in deep learning-based medical image segmentation. Recent studies use strong-weak pseudo supervision to leverage unlabeled data. However, performance is often hindered by inconsistencies between pseudo labels and their corresponding unlabeled images. In this work, we propose \textbf{SynMatch}, a novel framework that sidesteps the need for improving pseudo labels by synthesizing images to match them instead. Specifically, SynMatch synthesizes images using texture and shape features extracted from the same segmentation model that generates the corresponding pseudo labels for unlabeled images. This design enables the generation of highly consistent synthesized-image-pseudo-label pairs without requiring any training parameters for image synthesis. We extensively evaluate SynMatch across diverse medical image segmentation tasks under semi-supervised learning (SSL), weakly-supervised learning (WSL), and barely-supervised learning (BSL) settings with increasingly limited annotations. The results demonstrate that SynMatch achieves superior performance, especially in the most challenging BSL setting. For example, it outperforms the recent strong-weak pseudo supervision-based method by 29.71\% and 10.05\% on the polyp segmentation task with 5\% and 10\% scribble annotations, respectively. The code will be released at https://github.com/Senyh/SynMatch.

Zhiqiang Shen, Peng Cao, Jinzhu Yang et al.

Due to domain shifts across diverse medical imaging modalities, learned segmentation models often suffer significant performance degradation during deployment. We posit that these domain shifts can generally be categorized into two main components: 1) "style" shifts, referring to global disparities in image properties such as illumination, contrast, and color; and 2) "content" shifts, which involve local discrepancies in anatomical structures. To address the domain shifts in medical image segmentation, we first factorize an image into style codes and content maps, explicitly modeling the "style" and "content" components. Building on this, we introduce a Style-Content decomposition-based data augmentation algorithm (StyCona), which performs augmentation on both the global style and local content of source-domain images, enabling the training of a well-generalized model for domain generalizable medical image segmentation. StyCona is a simple yet effective plug-and-play module that substantially improves model generalization without requiring additional training parameters or modifications to segmentation model architectures. Experiments on cardiac magnetic resonance imaging and fundus photography segmentation tasks, with single and multiple target domains respectively, demonstrate the effectiveness of StyCona and its superiority over state-of-the-art domain generalization methods. The code is available at https://github.com/Senyh/StyCona.

Zhiqiang Shen, Peng Cao, Junming Su et al.

This paper investigates an extremely challenging problem: barely-supervised volumetric medical image segmentation (BSS). A BSS training dataset consists of two parts: 1) a barely-annotated labeled set, where each labeled image contains only a single-slice annotation, and 2) an unlabeled set comprising numerous unlabeled volumetric images. State-of-the-art BSS methods employ a registration-based paradigm, which uses inter-slice image registration to propagate single-slice annotations into volumetric pseudo labels, constructing a completely annotated labeled set, to which a semi-supervised segmentation scheme can be applied. However, the paradigm has a critical limitation: the pseudo-labels generated by image registration are unreliable and noisy. Motivated by this, we propose a new perspective: instead of solving BSS within a semi-supervised learning scheme, this work formulates BSS as an unsupervised domain adaptation problem. To this end, we propose a novel BSS framework, \textbf{B}arely-supervised learning \textbf{via} unsupervised domain \textbf{A}daptation (BvA), as an alternative to the dominant registration paradigm. Specifically, we first design a novel noise-free labeled data construction algorithm (NFC) for slice-to-volume labeled data synthesis. Then, we introduce a frequency and spatial Mix-Up strategy (FSX) to mitigate the domain shifts. Extensive experiments demonstrate that our method provides a promising alternative for BSS. Remarkably, the proposed method, trained on the left atrial segmentation dataset with \textbf{only one} barely-labeled image, achieves a Dice score of 81.20%, outperforming the state-of-the-art by 61.71%. The code is available at https://github.com/Senyh/BvA.

Haonan Wang, Peng Cao, Xiaoli Liu et al.

Most state-of-the-art methods for medical image segmentation adopt the encoder-decoder architecture. However, this U-shaped framework still has limitations in capturing the non-local multi-scale information with a simple skip connection. To solve the problem, we firstly explore the potential weakness of skip connections in U-Net on multiple segmentation tasks, and find that i) not all skip connections are useful, each skip connection has different contribution; ii) the optimal combinations of skip connections are different, relying on the specific datasets. Based on our findings, we propose a new segmentation framework, named UDTransNet, to solve three semantic gaps in U-Net. Specifically, we propose a Dual Attention Transformer (DAT) module for capturing the channel- and spatial-wise relationships to better fuse the encoder features, and a Decoder-guided Recalibration Attention (DRA) module for effectively connecting the DAT tokens and the decoder features to eliminate the inconsistency. Hence, both modules establish a learnable connection to solve the semantic gaps between the encoder and the decoder, which leads to a high-performance segmentation model for medical images. Comprehensive experimental results indicate that our UDTransNet produces higher evaluation scores and finer segmentation results with relatively fewer parameters over the state-of-the-art segmentation methods on different public datasets. Code: https://github.com/McGregorWwww/UDTransNet.

Haonan Wang, Peng Cao, Jiaqi Wang et al.

Most recent semantic segmentation methods adopt a U-Net framework with an encoder-decoder architecture. It is still challenging for U-Net with a simple skip connection scheme to model the global multi-scale context: 1) Not each skip connection setting is effective due to the issue of incompatible feature sets of encoder and decoder stage, even some skip connection negatively influence the segmentation performance; 2) The original U-Net is worse than the one without any skip connection on some datasets. Based on our findings, we propose a new segmentation framework, named UCTransNet (with a proposed CTrans module in U-Net), from the channel perspective with attention mechanism. Specifically, the CTrans module is an alternate of the U-Net skip connections, which consists of a sub-module to conduct the multi-scale Channel Cross fusion with Transformer (named CCT) and a sub-module Channel-wise Cross-Attention (named CCA) to guide the fused multi-scale channel-wise information to effectively connect to the decoder features for eliminating the ambiguity. Hence, the proposed connection consisting of the CCT and CCA is able to replace the original skip connection to solve the semantic gaps for an accurate automatic medical image segmentation. The experimental results suggest that our UCTransNet produces more precise segmentation performance and achieves consistent improvements over the state-of-the-art for semantic segmentation across different datasets and conventional architectures involving transformer or U-shaped framework. Code: https://github.com/McGregorWwww/UCTransNet.

Emre Aksan, Manuel Kaufmann, Peng Cao et al.

We propose a novel Transformer-based architecture for the task of generative modelling of 3D human motion. Previous work commonly relies on RNN-based models considering shorter forecast horizons reaching a stationary and often implausible state quickly. Recent studies show that implicit temporal representations in the frequency domain are also effective in making predictions for a predetermined horizon. Our focus lies on learning spatio-temporal representations autoregressively and hence generation of plausible future developments over both short and long term. The proposed model learns high dimensional embeddings for skeletal joints and how to compose a temporally coherent pose via a decoupled temporal and spatial self-attention mechanism. Our dual attention concept allows the model to access current and past information directly and to capture both the structural and the temporal dependencies explicitly. We show empirically that this effectively learns the underlying motion dynamics and reduces error accumulation over time observed in auto-regressive models. Our model is able to make accurate short-term predictions and generate plausible motion sequences over long horizons. We make our code publicly available at https://github.com/eth-ait/motion-transformer.

Yilun Xu, Peng Cao, Yuqing Kong et al.

Accurately annotating large scale dataset is notoriously expensive both in time and in money. Although acquiring low-quality-annotated dataset can be much cheaper, it often badly damages the performance of trained models when using such dataset without particular treatment. Various methods have been proposed for learning with noisy labels. However, most methods only handle limited kinds of noise patterns, require auxiliary information or steps (e.g. , knowing or estimating the noise transition matrix), or lack theoretical justification. In this paper, we propose a novel information-theoretic loss function, $\mathcal{L}_{DMI}$, for training deep neural networks robust to label noise. The core of $\mathcal{L}_{DMI}$ is a generalized version of mutual information, termed Determinant based Mutual Information (DMI), which is not only information-monotone but also relatively invariant. \emph{To the best of our knowledge, $\mathcal{L}_{DMI}$ is the first loss function that is provably robust to instance-independent label noise, regardless of noise pattern, and it can be applied to any existing classification neural networks straightforwardly without any auxiliary information}. In addition to theoretical justification, we also empirically show that using $\mathcal{L}_{DMI}$ outperforms all other counterparts in the classification task on both image dataset and natural language dataset include Fashion-MNIST, CIFAR-10, Dogs vs. Cats, MR with a variety of synthesized noise patterns and noise amounts, as well as a real-world dataset Clothing1M. Codes are available at https://github.com/Newbeeer/L_DMI .

Peng Cao, Yilun Xu, Yuqing Kong et al.

Eliciting labels from crowds is a potential way to obtain large labeled data. Despite a variety of methods developed for learning from crowds, a key challenge remains unsolved: \emph{learning from crowds without knowing the information structure among the crowds a priori, when some people of the crowds make highly correlated mistakes and some of them label effortlessly (e.g. randomly)}. We propose an information theoretic approach, Max-MIG, for joint learning from crowds, with a common assumption: the crowdsourced labels and the data are independent conditioning on the ground truth. Max-MIG simultaneously aggregates the crowdsourced labels and learns an accurate data classifier. Furthermore, we devise an accurate data-crowds forecaster that employs both the data and the crowdsourced labels to forecast the ground truth. To the best of our knowledge, this is the first algorithm that solves the aforementioned challenge of learning from crowds. In addition to the theoretical validation, we also empirically show that our algorithm achieves the new state-of-the-art results in most settings, including the real-world data, and is the first algorithm that is robust to various information structures. Codes are available at \hyperlink{https://github.com/Newbeeer/Max-MIG}{https://github.com/Newbeeer/Max-MIG}

Peng Cao, Zhijian Yang, Tennison Liu et al.

Wearable sensors enable the continuous acquisition of high-resolution physiological waveforms, such as photoplethysmography and accelerometry, under free-living conditions. However, inferring health-related phenotypes from these signals presents significant challenges due to high sampling frequencies, multimodal dependencies, and extreme sequence lengths (e.g., weeks of recordings), compounded by a scarcity of ground-truth labels. To address these challenges, existing self-supervised learning (SSL) methodologies typically follow two paradigms: (1) learning rich morphological representations from short waveform segments while collapsing longitudinal dynamics through simple aggregation, or (2) modeling behavioral patterns from coarse, hand-crafted features (e.g. heart rate, step counts) spanning longer horizons but foregoing subtle, predictive signatures in raw waveforms. To bridge this gap, we propose WavesFM, a foundation model utilizing a two-stage SSL framework for longitudinal physiological data. Specifically, we decompose the learning problem into two stages: first, a segment-level encoder is pretrained to extract local embeddings from short waveforms; subsequently, a temporal encoder is trained to model the sequence of these embeddings across a multi-day horizon. This hierarchical approach overcomes the computational complexity of high-resolution, long-sequence data, allowing the overall model to capture both local signal semantics and the complex circadian and inter-day variations governing physiological dynamics. Pretrained on over 6.8M hours (N=324k individuals) of recordings for the first stage and 5.3M hours (N=10k) for the second stage, WavesFM demonstrates superior performance across 58 diverse tasks spanning demographics, lifestyle, health conditions, and medications.

Jiarun Liu, Shiyue Xu, Yang Li et al.

Large Language Model agents face fundamental challenges in adapting to novel tasks due to limitations in tool availability and experience reuse. Existing approaches either rely on predefined tools with limited coverage or build tools from scratch without leveraging past experiences, leading to inefficient exploration and suboptimal performance. We introduce SMITH (Shared Memory Integrated Tool Hub), a unified cognitive architecture that seamlessly integrates dynamic tool creation with cross-task experience sharing through hierarchical memory organization. SMITH organizes agent memory into procedural, semantic, and episodic components, enabling systematic capability expansion while preserving successful execution patterns. Our approach formalizes tool creation as iterative code generation within controlled sandbox environments and experience sharing through episodic memory retrieval with semantic similarity matching. We further propose a curriculum learning strategy based on agent-ensemble difficulty re-estimation. Extensive experiments on the GAIA benchmark demonstrate SMITH's effectiveness, achieving 81.8% Pass@1 accuracy and outperforming state-of-the-art baselines including Alita (75.2%) and Memento (70.9%). Our work establishes a foundation for building truly adaptive agents that continuously evolve their capabilities through principled integration of tool creation and experience accumulation.

Ang Li, Qiugen Xiao, Peng Cao et al.

Reinforcement Learning from AI Feedback (RLAIF) has the advantages of shorter annotation cycles and lower costs over Reinforcement Learning from Human Feedback (RLHF), making it highly efficient during the rapid strategy iteration periods of large language model (LLM) training. Using ChatGPT as a labeler to provide feedback on open-domain prompts in RLAIF training, we observe an increase in human evaluators' preference win ratio for model responses, but a decrease in evaluators' satisfaction rate. Analysis suggests that the decrease in satisfaction rate is mainly due to some responses becoming less helpful, particularly in terms of correctness and truthfulness, highlighting practical limitations of basic RLAIF. In this paper, we propose Hybrid Reinforcement Learning from AI Feedback (HRLAIF). This method enhances the accuracy of AI annotations for responses, making the model's helpfulness more robust in training process. Additionally, it employs AI for Red Teaming, further improving the model's harmlessness. Human evaluation results show that HRLAIF inherits the ability of RLAIF to enhance human preference for outcomes at a low cost while also improving the satisfaction rate of responses. Compared to the policy model before Reinforcement Learning (RL), it achieves an increase of 2.08\% in satisfaction rate, effectively addressing the issue of a decrease of 4.58\% in satisfaction rate after basic RLAIF.

Qingshan Hou, Shuai Cheng, Peng Cao et al.

Representation learning offers a conduit to elucidate distinctive features within the latent space and interpret the deep models. However, the randomness of lesion distribution and the complexity of low-quality factors in medical images pose great challenges for models to extract key lesion features. Disease diagnosis methods guided by contrastive learning (CL) have shown significant advantages in lesion feature representation. Nevertheless, the effectiveness of CL is highly dependent on the quality of the positive and negative sample pairs. In this work, we propose a clinical-oriented multi-level CL framework that aims to enhance the model's capacity to extract lesion features and discriminate between lesion and low-quality factors, thereby enabling more accurate disease diagnosis from low-quality medical images. Specifically, we first construct multi-level positive and negative pairs to enhance the model's comprehensive recognition capability of lesion features by integrating information from different levels and qualities of medical images. Moreover, to improve the quality of the learned lesion embeddings, we introduce a dynamic hard sample mining method based on self-paced learning. The proposed CL framework is validated on two public medical image datasets, EyeQ and Chest X-ray, demonstrating superior performance compared to other state-of-the-art disease diagnostic methods.

Qingshan Hou, Meng Wang, Peng Cao et al.

Recent advancements in ophthalmology foundation models such as RetFound have demonstrated remarkable diagnostic capabilities but require massive datasets for effective pre-training, creating significant barriers for development and deployment. To address this critical challenge, we propose FundusGAN, a novel hierarchical feature-aware generative framework specifically designed for high-fidelity fundus image synthesis. Our approach leverages a Feature Pyramid Network within its encoder to comprehensively extract multi-scale information, capturing both large anatomical structures and subtle pathological features. The framework incorporates a modified StyleGAN-based generator with dilated convolutions and strategic upsampling adjustments to preserve critical retinal structures while enhancing pathological detail representation. Comprehensive evaluations on the DDR, DRIVE, and IDRiD datasets demonstrate that FundusGAN consistently outperforms state-of-the-art methods across multiple metrics (SSIM: 0.8863, FID: 54.2, KID: 0.0436 on DDR). Furthermore, disease classification experiments reveal that augmenting training data with FundusGAN-generated images significantly improves diagnostic accuracy across multiple CNN architectures (up to 6.49\% improvement with ResNet50). These results establish FundusGAN as a valuable foundation model component that effectively addresses data scarcity challenges in ophthalmological AI research, enabling more robust and generalizable diagnostic systems while reducing dependency on large-scale clinical data collection.

Zhi Liu, Tao Yang, Jing Wang et al.

Natural medicines, particularly Traditional Chinese Medicine (TCM), are gaining global recognition for their therapeutic potential in addressing human symptoms and diseases. TCM, with its systematic theories and extensive practical experience, provides abundant resources for healthcare. However, the effective application of TCM requires precise syndrome diagnosis, determination of treatment principles, and prescription formulation, which demand decades of clinical expertise. Despite advancements in TCM-based decision systems, machine learning, and deep learning research, limitations in data and single-objective constraints hinder their practical application. In recent years, large language models (LLMs) have demonstrated potential in complex tasks, but lack specialization in TCM and face significant challenges, such as too big model scale to deploy and issues with hallucination. To address these challenges, we introduce Tianyi with 7.6-billion-parameter LLM, a model scale proper and specifically designed for TCM, pre-trained and fine-tuned on diverse TCM corpora, including classical texts, expert treatises, clinical records, and knowledge graphs. Tianyi is designed to assimilate interconnected and systematic TCM knowledge through a progressive learning manner. Additionally, we establish TCMEval, a comprehensive evaluation benchmark, to assess LLMs in TCM examinations, clinical tasks, domain-specific question-answering, and real-world trials. The extensive evaluations demonstrate the significant potential of Tianyi as an AI assistant in TCM clinical practice and research, bridging the gap between TCM knowledge and practical application.

Yujian Liu, Linlang Cao, Chuang Chen et al.

Existing 3D head avatar reconstruction methods adopt a two-stage process, relying on tracked FLAME meshes derived from facial landmarks, followed by Gaussian-based rendering. However, misalignment between the estimated mesh and target images often leads to suboptimal rendering quality and loss of fine visual details. In this paper, we present MoGaFace, a novel 3D head avatar modeling framework that continuously refines facial geometry and texture attributes throughout the Gaussian rendering process. To address the misalignment between estimated FLAME meshes and target images, we introduce the Momentum-Guided Consistent Geometry module, which incorporates a momentum-updated expression bank and an expression-aware correction mechanism to ensure temporal and multi-view consistency. Additionally, we propose Latent Texture Attention, which encodes compact multi-view features into head-aware representations, enabling geometry-aware texture refinement via integration into Gaussians. Extensive experiments show that MoGaFace achieves high-fidelity head avatar reconstruction and significantly improves novel-view synthesis quality, even under inaccurate mesh initialization and unconstrained real-world settings.

Tianhong Li, Peng Cao, Yuan Yuan et al.

Real-world data often exhibits long tail distributions with heavy class imbalance, where the majority classes can dominate the training process and alter the decision boundaries of the minority classes. Recently, researchers have investigated the potential of supervised contrastive learning for long-tailed recognition, and demonstrated that it provides a strong performance gain. In this paper, we show that while supervised contrastive learning can help improve performance, past baselines suffer from poor uniformity brought in by imbalanced data distribution. This poor uniformity manifests in samples from the minority class having poor separability in the feature space. To address this problem, we propose targeted supervised contrastive learning (TSC), which improves the uniformity of the feature distribution on the hypersphere. TSC first generates a set of targets uniformly distributed on a hypersphere. It then makes the features of different classes converge to these distinct and uniformly distributed targets during training. This forces all classes, including minority classes, to maintain a uniform distribution in the feature space, improves class boundaries, and provides better generalization even in the presence of long-tail data. Experiments on multiple datasets show that TSC achieves state-of-the-art performance on long-tailed recognition tasks.

Xinwei Sun, Yilun Xu, Peng Cao et al.

Fusing data from multiple modalities provides more information to train machine learning systems. However, it is prohibitively expensive and time-consuming to label each modality with a large amount of data, which leads to a crucial problem of semi-supervised multi-modal learning. Existing methods suffer from either ineffective fusion across modalities or lack of theoretical guarantees under proper assumptions. In this paper, we propose a novel information-theoretic approach, namely \textbf{T}otal \textbf{C}orrelation \textbf{G}ain \textbf{M}aximization (TCGM), for semi-supervised multi-modal learning, which is endowed with promising properties: (i) it can utilize effectively the information across different modalities of unlabeled data points to facilitate training classifiers of each modality (ii) it has theoretical guarantee to identify Bayesian classifiers, i.e., the ground truth posteriors of all modalities. Specifically, by maximizing TC-induced loss (namely TC gain) over classifiers of all modalities, these classifiers can cooperatively discover the equivalent class of ground-truth classifiers; and identify the unique ones by leveraging limited percentage of labeled data. We apply our method to various tasks and achieve state-of-the-art results, including news classification, emotion recognition and disease prediction.

GuanXiong Luo, Na Zhao, Wenhao Jiang et al.

Purpose: To develop a deep learning-based Bayesian inference for MRI reconstruction. Methods: We modeled the MRI reconstruction problem with Bayes's theorem, following the recently proposed PixelCNN++ method. The image reconstruction from incomplete k-space measurement was obtained by maximizing the posterior possibility. A generative network was utilized as the image prior, which was computationally tractable, and the k-space data fidelity was enforced by using an equality constraint. The stochastic backpropagation was utilized to calculate the descent gradient in the process of maximum a posterior, and a projected subgradient method was used to impose the equality constraint. In contrast to the other deep learning reconstruction methods, the proposed one used the likelihood of prior as the training loss and the objective function in reconstruction to improve the image quality. Results: The proposed method showed an improved performance in preserving image details and reducing aliasing artifacts, compared with GRAPPA, $\ell_1$-ESPRiT, and MODL, a state-of-the-art deep learning reconstruction method. The proposed method generally achieved more than 5 dB peak signal-to-noise ratio improvement for compressed sensing and parallel imaging reconstructions compared with the other methods. Conclusion: The Bayesian inference significantly improved the reconstruction performance, compared with the conventional $\ell_1$-sparsity prior in compressed sensing reconstruction tasks. More importantly, the proposed reconstruction framework can be generalized for most MRI reconstruction scenarios.