Wenhui Zhu

Xiwen Chen, Wenhui Zhu, Jingjing Wang et al.

Aligning Large Language Models (LLMs) with human preferences is often formulated via Direct Preference Optimization (DPO). However, the standard Bradley-Terry instantiation of DPO is limited in modeling common departures from transitivity in human preferences. To address this, recent work has introduced Self-Play Preference Optimization (SPPO), which iteratively refines the policy by training on self-generated win-lose pairs. Our investigation, however, reveals a critical instability in SPPO: the optimization is prone to policy degeneration when the preference oracle assigns overly confident wins to semantically indistinguishable responses. To mitigate this, we propose S-SPPO, a dual-space semantic calibration framework comprising: i) Supervision Calibration via semantic gating, which anneals win rate targets toward the maximum-entropy baseline as semantic overlap increases; and ii) Representation Calibration via latent repulsion to enforce geometric diversity to prevent manifold collapse and maintain latent diversity between chosen and rejected samples. Theoretically, we show that the calibration preserves the constant-sum game structure, facilitating convergence to a Nash Equilibrium. Empirically, S-SPPO avoids the performance degradation seen in prior methods, achieving 52.19% win rate and 47.46% length-controlled win rate on AlpacaEval 2.0 with Llama-3-8B, without using additional human-annotated preferences during training. The code will be available at https://github.com/xiwenc1/s-sppo.

Wenhui Zhu, Xiwen Chen, Peijie Qiu et al.

Multiple instance learning (MIL) stands as a powerful approach in weakly supervised learning, regularly employed in histological whole slide image (WSI) classification for detecting tumorous lesions. However, existing mainstream MIL methods focus on modeling correlation between instances while overlooking the inherent diversity among instances. However, few MIL methods have aimed at diversity modeling, which empirically show inferior performance but with a high computational cost. To bridge this gap, we propose a novel MIL aggregation method based on diverse global representation (DGR-MIL), by modeling diversity among instances through a set of global vectors that serve as a summary of all instances. First, we turn the instance correlation into the similarity between instance embeddings and the predefined global vectors through a cross-attention mechanism. This stems from the fact that similar instance embeddings typically would result in a higher correlation with a certain global vector. Second, we propose two mechanisms to enforce the diversity among the global vectors to be more descriptive of the entire bag: (i) positive instance alignment and (ii) a novel, efficient, and theoretically guaranteed diversification learning paradigm. Specifically, the positive instance alignment module encourages the global vectors to align with the center of positive instances (e.g., instances containing tumors in WSI). To further diversify the global representations, we propose a novel diversification learning paradigm leveraging the determinantal point process. The proposed model outperforms the state-of-the-art MIL aggregation models by a substantial margin on the CAMELYON-16 and the TCGA-lung cancer datasets. The code is available at \url{https://github.com/ChongQingNoSubway/DGR-MIL}.

Xuanzhao Dong, Wenhui Zhu, Peijie Qiu et al.

Despite their popularity and success, Multimodal Large Language Models (MLLMs) often struggle to interpret images accurately, which limits their reasoning capability in complex scenarios (e.g., high object density and complex background clutter). Prior work mainly addresses this limitation by incorporating explicit visual cues like bounding boxes that require extra annotations. In addition, the resulting low-resolution crops often miss fine-grained details that MLLMs require for accurate reasoning. Therefore, we propose Mags-RL, an Agentic Reinforcement Learning (RL) framework that equips MLLMs with an external super-resolution "magnifying glass" agent for high-resolution fine-grained inspection. Specifically, the model performs two-round reasoning: in the first round, it generates an initial rationale and autonomously identifies regions of interest without relying on additional annotations; in the second round, it invokes a super-resolution agent to crop and upscale those regions, then revisits and verifies its earlier reasoning to produce the final answer. We also introduce a novel curriculum learning strategy that enables data-efficient RL training, needing as few as only 40 training samples to achieve reasonable performance. Experiments on VSR, TallyQA, and GQA subsets show its superior performance against recent strong competing methods, demonstrating high-quality reasoning with precise visual grounding. Code and weights will be released soon.

Peijie Qiu, Pan Xiao, Wenhui Zhu et al.

Multiple Instance Learning (MIL) has been widely used in weakly supervised whole slide image (WSI) classification. Typical MIL methods include a feature embedding part, which embeds the instances into features via a pre-trained feature extractor, and an MIL aggregator that combines instance embeddings into predictions. Most efforts have typically focused on improving these parts. This involves refining the feature embeddings through self-supervised pre-training as well as modeling the correlations between instances separately. In this paper, we proposed a sparsely coding MIL (SC-MIL) method that addresses those two aspects at the same time by leveraging sparse dictionary learning. The sparse dictionary learning captures the similarities of instances by expressing them as sparse linear combinations of atoms in an over-complete dictionary. In addition, imposing sparsity improves instance feature embeddings by suppressing irrelevant instances while retaining the most relevant ones. To make the conventional sparse coding algorithm compatible with deep learning, we unrolled it into a sparsely coded module leveraging deep unrolling. The proposed SC module can be incorporated into any existing MIL framework in a plug-and-play manner with an acceptable computational cost. The experimental results on multiple datasets demonstrated that the proposed SC module could substantially boost the performance of state-of-the-art MIL methods. The codes are available at \href{https://github.com/sotiraslab/SCMIL.git}{https://github.com/sotiraslab/SCMIL.git}.

Wenhui Zhu, Peijie Qiu, Xiwen Chen et al.

Over the past few decades, convolutional neural networks (CNNs) have been at the forefront of the detection and tracking of various retinal diseases (RD). Despite their success, the emergence of vision transformers (ViT) in the 2020s has shifted the trajectory of RD model development. The leading-edge performance of ViT-based models in RD can be largely credited to their scalability-their ability to improve as more parameters are added. As a result, ViT-based models tend to outshine traditional CNNs in RD applications, albeit at the cost of increased data and computational demands. ViTs also differ from CNNs in their approach to processing images, working with patches rather than local regions, which can complicate the precise localization of small, variably presented lesions in RD. In our study, we revisited and updated the architecture of a CNN model, specifically MobileNet, to enhance its utility in RD diagnostics. We found that an optimized MobileNet, through selective modifications, can surpass ViT-based models in various RD benchmarks, including diabetic retinopathy grading, detection of multiple fundus diseases, and classification of diabetic macular edema. The code is available at https://github.com/Retinal-Research/NN-MOBILENET

Vamsi Krishna Vasa, Peijie Qiu, Wenhui Zhu et al.

Retinal fundus photography offers a non-invasive way to diagnose and monitor a variety of retinal diseases, but is prone to inherent quality glitches arising from systemic imperfections or operator/patient-related factors. However, high-quality retinal images are crucial for carrying out accurate diagnoses and automated analyses. The fundus image enhancement is typically formulated as a distribution alignment problem, by finding a one-to-one mapping between a low-quality image and its high-quality counterpart. This paper proposes a context-informed optimal transport (OT) learning framework for tackling unpaired fundus image enhancement. In contrast to standard generative image enhancement methods, which struggle with handling contextual information (e.g., over-tampered local structures and unwanted artifacts), the proposed context-aware OT learning paradigm better preserves local structures and minimizes unwanted artifacts. Leveraging deep contextual features, we derive the proposed context-aware OT using the earth mover's distance and show that the proposed context-OT has a solid theoretical guarantee. Experimental results on a large-scale dataset demonstrate the superiority of the proposed method over several state-of-the-art supervised and unsupervised methods in terms of signal-to-noise ratio, structural similarity index, as well as two downstream tasks. The code is available at \url{https://github.com/Retinal-Research/Contextual-OT}.

Xuanzhao Dong, Wenhui Zhu, Xiwen Chen et al.

The advancement of general medical Multimodal Large Language Models (MLLMs) has shown great potential for building conversational assistants to support clinical diagnosis. However, their adaptation to highly specialized domains such as ophthalmology remains underexplored, primarily due to the scarcity of large-scale, domain-specific instruction-tuning data. Existing ophthalmic datasets for conversational agents are often limited in scale and largely rely on images from established public benchmarks, limiting the scalability of ophthalmic MLLMs and their ability to capture real-world clinical complexity. To address this gap, we propose $\textbf{OphIn-Engine}$, an ophthalmology-specific instruction data curation pipeline that constructs high-quality instruction data from open-access ophthalmology web-scale videos. The pipeline integrates multimodal transcription for extracting image-transcript pairs, visual cue separation and scoring for identifying clinically relevant visual descriptions, and instruction synthesis with quality control for generating accurate and diverse clinical dialogues. Using this engine, we introduce $\textbf{OphIn-500K}$, a large-scale multimodal ophthalmology instruction-tuning dataset containing over 500,000 instruction instances and more than 151,000 unique images from over 29,000 video clips, formatted as visual question answering (VQA), multi-turn conversational interactions, and chain-of-thought (CoT) reasoning. Built upon this dataset, we further develop $\textbf{OphIn-VL}$, an ophthalmology-specific MLLM with advanced visual understanding and conversational capabilities. Comprehensive experiments and case studies demonstrate that OphIn-VL achieves superior performance compared with state-of-the-art general medical and domain-specific MLLMs.

Hao Wang, Wenhui Zhu, Jiayou Qin et al.

Detecting retinal image analysis, particularly the geometrical features of branching points, plays an essential role in diagnosing eye diseases. However, existing methods used for this purpose often are coarse-level and lack fine-grained analysis for efficient annotation. To mitigate these issues, this paper proposes a novel method for detecting retinal branching angles using a self-configured image processing technique. Additionally, we offer an open-source annotation tool and a benchmark dataset comprising 40 images annotated with retinal branching angles. Our methodology for retinal branching angle detection and calculation is detailed, followed by a benchmark analysis comparing our method with previous approaches. The results indicate that our method is robust under various conditions with high accuracy and efficiency, which offers a valuable instrument for ophthalmic research and clinical applications.

Xuanzhao Dong, Vamsi Krishna Vasa, Wenhui Zhu et al.

Retinal fundus photography is significant in diagnosing and monitoring retinal diseases. However, systemic imperfections and operator/patient-related factors can hinder the acquisition of high-quality retinal images. Previous efforts in retinal image enhancement primarily relied on GANs, which are limited by the trade-off between training stability and output diversity. In contrast, the Schrödinger Bridge (SB), offers a more stable solution by utilizing Optimal Transport (OT) theory to model a stochastic differential equation (SDE) between two arbitrary distributions. This allows SB to effectively transform low-quality retinal images into their high-quality counterparts. In this work, we leverage the SB framework to propose an image-to-image translation pipeline for retinal image enhancement. Additionally, previous methods often fail to capture fine structural details, such as blood vessels. To address this, we enhance our pipeline by introducing Dynamic Snake Convolution, whose tortuous receptive field can better preserve tubular structures. We name the resulting retinal fundus image enhancement framework the Context-aware Unpaired Neural Schrödinger Bridge (CUNSB-RFIE). To the best of our knowledge, this is the first endeavor to use the SB approach for retinal image enhancement. Experimental results on a large-scale dataset demonstrate the advantage of the proposed method compared to several state-of-the-art supervised and unsupervised methods in terms of image quality and performance on downstream tasks.The code is available at https://github.com/Retinal-Research/CUNSB-RFIE .

Jianfeng Wu, Yi Su, Yanxi Chen et al.

Background: Alzheimer's Disease (AD) is the most common type of age-related dementia, affecting 6.2 million people aged 65 or older according to CDC data. It is commonly agreed that discovering an effective AD diagnosis biomarker could have enormous public health benefits, potentially preventing or delaying up to 40% of dementia cases. Tau neurofibrillary tangles are the primary driver of downstream neurodegeneration and subsequent cognitive impairment in AD, resulting in structural deformations such as hippocampal atrophy that can be observed in magnetic resonance imaging (MRI) scans. Objective: To build a surface-based model to 1) detect differences between APOE subgroups in patterns of tau deposition and hippocampal atrophy, and 2) use the extracted surface-based features to predict cognitive decline. Methods: Using data obtained from different institutions, we develop a surface-based federated Chow test model to study the synergistic effects of APOE, a previously reported significant risk factor of AD, and tau on hippocampal surface morphometry. Results: We illustrate that the APOE-specific morphometry features correlate with AD progression and better predict future AD conversion than other MRI biomarkers. For example, a strong association between atrophy and abnormal tau was identified in hippocampal subregion cornu ammonis 1 (CA1 subfield) and subiculum in e4 homozygote cohort. Conclusion: Our model allows for identifying MRI biomarkers for AD and cognitive decline prediction and may uncover a corner of the neural mechanism of the influence of APOE and tau deposition on hippocampal morphology.

Wenhui Zhu, Xuanzhao Dong, Xiwen Chen et al.

The rapid deployment of open-source frameworks has significantly advanced the development of modern multi-agent systems. However, expanded action spaces, including uncontrolled privilege exposure and hidden inter-system interactions, pose severe security challenges. Specifically, Indirect Prompt Injections (IPI), which conceal malicious instructions within third-party content, can trigger unauthorized actions such as data exfiltration during normal operations. While current security evaluations predominantly rely on isolated single-turn benchmarks, the systemic vulnerabilities of these agents within complex dynamic environments remain critically underexplored. To bridge this gap, we systematically evaluate six defense strategies against four sophisticated IPI attack vectors across nine LLM backbones. Crucially, we conduct our evaluation entirely within dynamic multi-step tool-calling environments to capture the true attack surface of modern autonomous agents. Moving beyond binary success rates, our multidimensional analysis reveals a pronounced fragility. Advanced injections successfully bypass nearly all baseline defenses, and some surface-level mitigations even produce counterproductive side effects. Furthermore, while agents execute malicious instructions almost instantaneously, their internal states exhibit abnormally high decision entropy. Motivated by this latent hesitation, we investigate Representation Engineering (RepE) as a robust detection strategy. By extracting hidden states at the tool-input position, we revealed that the RepE-based circuit breaker successfully identifies and intercepts unauthorized actions before the agent commits to them, achieving high detection accuracy across diverse LLM backbones. This study exposes the limitations of current IPI defenses and provides a highly practical paradigm for building resilient multi-agent architectures.

Jianfeng Wu, Yi Su, Wenhui Zhu et al.

Background: Beta-amyloid (A$β$) plaques and tau protein tangles in the brain are the defining 'A' and 'T' hallmarks of Alzheimer's disease (AD), and together with structural atrophy detectable on brain magnetic resonance imaging (MRI) scans as one of the neurodegenerative ('N') biomarkers comprise the ''ATN framework'' of AD. Current methods to detect A$β$/tau pathology include cerebrospinal fluid (CSF; invasive), positron emission tomography (PET; costly and not widely available), and blood-based biomarkers (BBBM; promising but mainly still in development). Objective: To develop a non-invasive and widely available structural MRI-based framework to quantitatively predict the amyloid and tau measurements. Methods: With MRI-based hippocampal multivariate morphometry statistics (MMS) features, we apply our Patch Analysis-based Surface Correntropy-induced Sparse coding and max-pooling (PASCS-MP) method combined with the ridge regression model to individual amyloid/tau measure prediction. Results: We evaluate our framework on amyloid PET/MRI and tau PET/MRI datasets from the Alzheimer's Disease Neuroimaging Initiative (ADNI). Each subject has one pair consisting of a PET image and MRI scan, collected at about the same time. Experimental results suggest that amyloid/tau measurements predicted with our PASCP-MP representations are closer to the real values than the measures derived from other approaches, such as hippocampal surface area, volume, and shape morphometry features based on spherical harmonics (SPHARM). Conclusion: The MMS-based PASCP-MP is an efficient tool that can bridge hippocampal atrophy with amyloid and tau pathology and thus help assess disease burden, progression, and treatment effects.

Wenhui Zhu, Peijie Qiu, Natasha Lepore et al.

Lesion appearance is a crucial clue for medical providers to distinguish referable diabetic retinopathy (rDR) from non-referable DR. Most existing large-scale DR datasets contain only image-level labels rather than pixel-based annotations. This motivates us to develop algorithms to classify rDR and segment lesions via image-level labels. This paper leverages self-supervised equivariant learning and attention-based multi-instance learning (MIL) to tackle this problem. MIL is an effective strategy to differentiate positive and negative instances, helping us discard background regions (negative instances) while localizing lesion regions (positive ones). However, MIL only provides coarse lesion localization and cannot distinguish lesions located across adjacent patches. Conversely, a self-supervised equivariant attention mechanism (SEAM) generates a segmentation-level class activation map (CAM) that can guide patch extraction of lesions more accurately. Our work aims at integrating both methods to improve rDR classification accuracy. We conduct extensive validation experiments on the Eyepacs dataset, achieving an area under the receiver operating characteristic curve (AU ROC) of 0.958, outperforming current state-of-the-art algorithms.

Wenhui Zhu, Xiwen Chen, Zhipeng Wang et al.

Long-horizon LLM agents require memory systems that remain accurate under fixed context budgets. However, existing systems struggle with two persistent challenges in long-term dialogue: (i) \textbf{disconnected evidence}, where multi-hop answers require linking facts distributed across time, and (ii) \textbf{state updates}, where evolving information (e.g., schedule changes) creates conflicts with older static logs. We propose AriadneMem, a structured memory system that addresses these failure modes via a decoupled two-phase pipeline. In the \textbf{offline construction phase}, AriadneMem employs \emph{entropy-aware gating} to filter noise and low-information message before LLM extraction and applies \emph{conflict-aware coarsening} to merge static duplicates while preserving state transitions as temporal edges. In the \textbf{online reasoning phase}, rather than relying on expensive iterative planning, AriadneMem executes \emph{algorithmic bridge discovery} to reconstruct missing logical paths between retrieved facts, followed by \emph{single-call topology-aware synthesis}. On LoCoMo experiments with GPT-4o, AriadneMem improves \textbf{Multi-Hop F1 by 15.2\%} and \textbf{Average F1 by 9.0\%} over strong baselines. Crucially, by offloading reasoning to the graph layer, AriadneMem reduces \textbf{total runtime by 77.8\%} using only \textbf{497} context tokens. The code is available at https://github.com/LLM-VLM-GSL/AriadneMem.

Mohammad Farazi, Zhangsihao Yang, Wenhui Zhu et al.

Convolutional neural networks (CNN) have been broadly studied on images, videos, graphs, and triangular meshes. However, it has seldom been studied on tetrahedral meshes. Given the merits of using volumetric meshes in applications like brain image analysis, we introduce a novel interpretable graph CNN framework for the tetrahedral mesh structure. Inspired by ChebyNet, our model exploits the volumetric Laplace-Beltrami Operator (LBO) to define filters over commonly used graph Laplacian which lacks the Riemannian metric information of 3D manifolds. For pooling adaptation, we introduce new objective functions for localized minimum cuts in the Graclus algorithm based on the LBO. We employ a piece-wise constant approximation scheme that uses the clustering assignment matrix to estimate the LBO on sampled meshes after each pooling. Finally, adapting the Gradient-weighted Class Activation Mapping algorithm for tetrahedral meshes, we use the obtained heatmaps to visualize discovered regions-of-interest as biomarkers. We demonstrate the effectiveness of our model on cortical tetrahedral meshes from patients with Alzheimer's disease, as there is scientific evidence showing the correlation of cortical thickness to neurodegenerative disease progression. Our results show the superiority of our LBO-based convolution layer and adapted pooling over the conventionally used unitary cortical thickness, graph Laplacian, and point cloud representation.

Mohammad Farazi, Wenhui Zhu, Zhangsihao Yang et al.

This paper studies 3D dense shape correspondence, a key shape analysis application in computer vision and graphics. We introduce a novel hybrid geometric deep learning-based model that learns geometrically meaningful and discretization-independent features with a U-Net model as the primary node feature extraction module, followed by a successive spectral-based graph convolutional network. To create a diverse set of filters, we use anisotropic wavelet basis filters, being sensitive to both different directions and band-passes. This filter set overcomes the over-smoothing behavior of conventional graph neural networks. To further improve the model's performance, we add a function that perturbs the feature maps in the last layer ahead of fully connected layers, forcing the network to learn more discriminative features overall. The resulting correspondence maps show state-of-the-art performance on the benchmark datasets based on average geodesic errors and superior robustness to discretization in 3D meshes. Our approach provides new insights and practical solutions to the dense shape correspondence research.

Yanxi Chen, Wenhui Zhu, Xiwen Chen et al.

Although Large Audio-Language Models (LALMs) deliver state-of-the-art (SOTA) performance, they frequently suffer from hallucinations, e.g. generating text not grounded in the audio input. We analyze these grounding failures and identify a distinct taxonomy: Event Omission, False Event Identity, Temporal Relation Error, and Quantitative Temporal Error. To address this, we introduce the AHA (Audio Hallucination Alignment) framework. By leveraging counterfactual hard negative mining, our pipeline constructs a high-quality preference dataset that forces models to distinguish strict acoustic evidence from linguistically plausible fabrications. Additionally, we establish AHA-Eval, a diagnostic benchmark designed to rigorously test these fine-grained temporal reasoning capabilities. We apply this data to align Qwen2.5-Omni. The resulting model, Qwen-Audio-AHA, achieves a 13.7% improvement on AHA-Eval. Crucially, this benefit generalizes beyond our diagnostic set. Our model shows substantial gains on public benchmarks, including 1.3% on MMAU-Test and 1.6% on MMAR, outperforming latest SOTA methods. The model and dataset are open-sourced at https://github.com/LLM-VLM-GSL/AHA.

Wenhui Zhu, Peijie Qiu, Mohammad Farazi et al.

Real-world non-mydriatic retinal fundus photography is prone to artifacts, imperfections and low-quality when certain ocular or systemic co-morbidities exist. Artifacts may result in inaccuracy or ambiguity in clinical diagnoses. In this paper, we proposed a simple but effective end-to-end framework for enhancing poor-quality retinal fundus images. Leveraging the optimal transport theory, we proposed an unpaired image-to-image translation scheme for transporting low-quality images to their high-quality counterparts. We theoretically proved that a Generative Adversarial Networks (GAN) model with a generator and discriminator is sufficient for this task. Furthermore, to mitigate the inconsistency of information between the low-quality images and their enhancements, an information consistency mechanism was proposed to maximally maintain structural consistency (optical discs, blood vessels, lesions) between the source and enhanced domains. Extensive experiments were conducted on the EyeQ dataset to demonstrate the superiority of our proposed method perceptually and quantitatively.

Wenhui Zhu, Peijie Qiu, Oana M. Dumitrascu et al.

Non-mydriatic retinal color fundus photography (CFP) is widely available due to the advantage of not requiring pupillary dilation, however, is prone to poor quality due to operators, systemic imperfections, or patient-related causes. Optimal retinal image quality is mandated for accurate medical diagnoses and automated analyses. Herein, we leveraged the Optimal Transport (OT) theory to propose an unpaired image-to-image translation scheme for mapping low-quality retinal CFPs to high-quality counterparts. Furthermore, to improve the flexibility, robustness, and applicability of our image enhancement pipeline in the clinical practice, we generalized a state-of-the-art model-based image reconstruction method, regularization by denoising, by plugging in priors learned by our OT-guided image-to-image translation network. We named it as regularization by enhancing (RE). We validated the integrated framework, OTRE, on three publicly available retinal image datasets by assessing the quality after enhancement and their performance on various downstream tasks, including diabetic retinopathy grading, vessel segmentation, and diabetic lesion segmentation. The experimental results demonstrated the superiority of our proposed framework over some state-of-the-art unsupervised competitors and a state-of-the-art supervised method.

Jelena Markovic-Voronov, Wenhui Zhu, Bo Long et al.

We introduce a principled probabilistic framework for reward-guided decoding in large language models, addressing the limitations of standard decoding methods that optimize token-level likelihood rather than sequence-level quality. Our method defines a reward-augmented target distribution over complete sequences by combining model transition probabilities with prefix-dependent reward potentials. Importantly, the approach is training-free: it leaves model weights unchanged and instead modifies the inference distribution via reward potentials, with all gains arising purely from inference-time sampling. To sample from this distribution, we develop Sequential Monte Carlo algorithms, including a computationally efficient prefix-only variant and a lookahead variant whose intermediate targets match the exact marginals of the full sequence distribution. The framework also integrates resample-move updates with Metropolis-Hastings rejuvenation and supports block-wise generation, subsuming common decoding strategies such as temperature sampling and power-tempered objectives. Empirical results across three 7B models show significant gains. On code generation (HumanEval), our method improves base performance by up to 54.9% and surpasses the strongest sampling baselines by 9.1%-15.3%. On mathematical reasoning (MATH500), it achieves gains of up to 8.8%. Notably, it reaches 87.8% on HumanEval and 78.4% on MATH500 with Qwen2.5-7B, consistently outperforming the reinforcement learning method GRPO.

Xiwen Chen, Jingjing Wang, Wenhui Zhu et al.

Black-box knowledge distillation for large language models presents a strict trade-off. Simple off-policy methods (e.g., sequence-level knowledge distillation) struggle to correct the student's inherent errors. Fully on-policy methods (e.g., Generative Adversarial Distillation) solve this via adversarial training but introduce well-known training instability and crippling computational overhead. To address this dilemma, we propose SODA (Semi On-policy Distillation with Alignment), a highly efficient alternative motivated by the inherent capability gap between frontier teachers and much smaller base models. Because a compact student model's natural, zero-shot responses are almost strictly inferior to the powerful teacher's targets, we can construct a highly effective contrastive signal simply by pairing the teacher's optimal response with a one-time static snapshot of the student's outputs. This demonstrates that exposing the small student to its own static inferior behaviors is sufficient for high-quality distribution alignment, eliminating the need for costly dynamic rollouts and fragile adversarial balancing. Extensive evaluations across four compact Qwen2.5 and Llama-3 models validate this semi on-policy paradigm. SODA matches or outperforms the state-of-the-art methods on 15 out of 16 benchmark results. More importantly, it achieves this superior distillation quality while training 10 times faster, consuming 27% less peak GPU memory, and completely eliminating adversarial instability.

Wenhui Zhu, Peijie Qiu, Xiwen Chen et al.

Multiple instance learning (MIL) was a weakly supervised learning approach that sought to assign binary class labels to collections of instances known as bags. However, due to their weak supervision nature, the MIL methods were susceptible to overfitting and required assistance in developing comprehensive representations of target instances. While regularization typically effectively combated overfitting, its integration with the MIL model has been frequently overlooked in prior studies. Meanwhile, current regularization methods for MIL have shown limitations in their capacity to uncover a diverse array of representations. In this study, we delve into the realm of regularization within the MIL model, presenting a novel approach in the form of a Progressive Dropout Layer (PDL). We aim to not only address overfitting but also empower the MIL model in uncovering intricate and impactful feature representations. The proposed method was orthogonal to existing MIL methods and could be easily integrated into them to boost performance. Our extensive evaluation across a range of MIL benchmark datasets demonstrated that the incorporation of the PDL into multiple MIL methods not only elevated their classification performance but also augmented their potential for weakly-supervised feature localizations.

Rui Cai, Weijie Jacky Mo, Xiaofei Wen et al.

The open-source model ecosystem now contains hundreds of thousands of pretrained models, yet picking the best model for a new dataset is increasingly infeasible: new models and unbenchmarked datasets emerge continuously, leaving practitioners with no prior records on either side. Existing approaches handle only fragments of this in-the-wild setting: AutoML and transferability estimation select models from small predefined pools or require expensive per-model forward passes on the target dataset, while model routing presupposes a given candidate pool. We introduce ModelLens, a unified framework for model recommendation in the wild. Our key insight is that public leaderboard interactions, though scattered and noisy, collectively trace out an implicit atlas of model capabilities across heterogeneous evaluation settings, a signal rich enough to learn from directly. By learning a performance-aware latent space over model--dataset--metric tuples, ModelLens ranks unseen models on unseen datasets without running candidates on the target dataset. On a new benchmark of 1.62M evaluation records spanning 47K models and 9.6K datasets, ModelLens surpasses baselines that either rely on metadata alone or require running each candidate on the target dataset. Its recommended Top-K pools further improve multiple representative routing methods by up to 81% across diverse QA benchmarks. Case studies on recently released benchmarks further confirm generalization to both text and vision-language tasks.

Xin Li, Wenhui Zhu, Xuanzhao Dong et al.

Retinal imaging is a critical, non-invasive modality for the early detection and monitoring of ocular and systemic diseases. Deep learning, particularly convolutional neural networks (CNNs), has significant progress in automated retinal analysis, supporting tasks such as fundus image classification, lesion detection, and vessel segmentation. As a representative lightweight network, nnMobileNet has demonstrated strong performance across multiple retinal benchmarks while remaining computationally efficient. However, purely convolutional architectures inherently struggle to capture long-range dependencies and model the irregular lesions and elongated vascular patterns that characterize on retinal images, despite the critical importance of vascular features for reliable clinical diagnosis. To further advance this line of work and extend the original vision of nnMobileNet, we propose nnMobileNet++, a hybrid architecture that progressively bridges convolutional and transformer representations. The framework integrates three key components: (i) dynamic snake convolution for boundary-aware feature extraction, (ii) stage-specific transformer blocks introduced after the second down-sampling stage for global context modeling, and (iii) retinal image pretraining to improve generalization. Experiments on multiple public retinal datasets for classification, together with ablation studies, demonstrate that nnMobileNet++ achieves state-of-the-art or highly competitive accuracy while maintaining low computational cost, underscoring its potential as a lightweight yet effective framework for retinal image analysis.

Xiwen Chen, Wenhui Zhu, Peijie Qiu et al.

Recent advances in reinforcement learning for language model post-training, such as Group Relative Policy Optimization (GRPO), have shown promise in low-resource settings. However, GRPO typically relies on solution-level and scalar reward signals that fail to capture the semantic diversity among sampled completions. This leads to what we identify as a diversity-quality inconsistency, where distinct reasoning paths may receive indistinguishable rewards. To address this limitation, we propose $\textit{Diversity-aware Reward Adjustment}$ (DRA), a method that explicitly incorporates semantic diversity into the reward computation. DRA uses Submodular Mutual Information (SMI) to downweight redundant completions and amplify rewards for diverse ones. This encourages better exploration during learning, while maintaining stable exploitation of high-quality samples. Our method integrates seamlessly with both GRPO and its variant DR.~GRPO, resulting in $\textit{DRA-GRPO}$ and $\textit{DGA-DR.~GRPO}$. We evaluate our method on five mathematical reasoning benchmarks and find that it outperforms recent strong baselines. It achieves state-of-the-art performance with an average accuracy of 58.2%, using only 7,000 fine-tuning samples and a total training cost of approximately $55. The code is available at https://github.com/xiwenc1/DRA-GRPO.

Wenhui Zhu, Xin Li, Xiwen Chen et al.

Recently, Multimodal Large Language Models (MLLMs) have gained significant attention for their remarkable ability to process and analyze non-textual data, such as images, videos, and audio. Notably, several adaptations of general-domain MLLMs to the medical field have been explored, including LLaVA-Med. However, these medical adaptations remain insufficiently advanced in understanding and interpreting retinal images. In contrast, medical experts emphasize the importance of quantitative analyses for disease detection and interpretation. This underscores a gap between general-domain and medical-domain MLLMs: while general-domain MLLMs excel in broad applications, they lack the specialized knowledge necessary for precise diagnostic and interpretative tasks in the medical field. To address these challenges, we introduce \textit{RetinalGPT}, a multimodal conversational assistant for clinically preferred quantitative analysis of retinal images. Specifically, we achieve this by compiling a large retinal image dataset, developing a novel data pipeline, and employing customized visual instruction tuning to enhance both retinal analysis and enrich medical knowledge. In particular, RetinalGPT outperforms MLLM in the generic domain by a large margin in the diagnosis of retinal diseases in 8 benchmark retinal datasets. Beyond disease diagnosis, RetinalGPT features quantitative analyses and lesion localization, representing a pioneering step in leveraging LLMs for an interpretable and end-to-end clinical research framework. The code is available at https://github.com/Retinal-Research/RetinalGPT

Zhangsihao Yang, Mengyi Shan, Mohammad Farazi et al.

Human video generation task has gained significant attention with the advancement of deep generative models. Generating realistic videos with human movements is challenging in nature, due to the intricacies of human body topology and sensitivity to visual artifacts. The extensively studied 2D media generation methods take advantage of massive human media datasets, but struggle with 3D-aware control; whereas 3D avatar-based approaches, while offering more freedom in control, lack photorealism and cannot be harmonized seamlessly with background scene. We propose AMG, a method that combines the 2D photorealism and 3D controllability by conditioning video diffusion models on controlled rendering of 3D avatars. We additionally introduce a novel data processing pipeline that reconstructs and renders human avatar movements from dynamic camera videos. AMG is the first method that enables multi-person diffusion video generation with precise control over camera positions, human motions, and background style. We also demonstrate through extensive evaluation that it outperforms existing human video generation methods conditioned on pose sequences or driving videos in terms of realism and adaptability.

Xuanzhao Dong, Wenhui Zhu, Hao Wang et al.

Medical question answering (QA) is a reasoning-intensive task that remains challenging for large language models (LLMs) due to hallucinations and outdated domain knowledge. Retrieval-Augmented Generation (RAG) provides a promising post-training solution by leveraging external knowledge. However, existing medical RAG systems suffer from two key limitations: (1) a lack of modeling for human-like reasoning behaviors during information retrieval, and (2) reliance on suboptimal medical corpora, which often results in the retrieval of irrelevant or noisy snippets. To overcome these challenges, we propose Discuss-RAG, a plug-and-play module designed to enhance the medical QA RAG system through collaborative agent-based reasoning. Our method introduces a summarizer agent that orchestrates a team of medical experts to emulate multi-turn brainstorming, thereby improving the relevance of retrieved content. Additionally, a decision-making agent evaluates the retrieved snippets before their final integration. Experimental results on four benchmark medical QA datasets show that Discuss-RAG consistently outperforms MedRAG, especially significantly improving answer accuracy by up to 16.67% on BioASQ and 12.20% on PubMedQA. The code is available at: https://github.com/LLM-VLM-GSL/Discuss-RAG.

Xuanzhao Dong, Wenhui Zhu, Xiwen Chen et al.

Autoregressive models (ARMs) have long dominated the landscape of biomedical vision-language models (VLMs). Recently, masked diffusion models such as LLaDA have emerged as promising alternatives, yet their application in the biomedical domain remains largely underexplored. To bridge this gap, we introduce \textbf{LLaDA-MedV}, the first large language diffusion model tailored for biomedical image understanding through vision instruction tuning. LLaDA-MedV achieves relative performance gains of 7.855\% over LLaVA-Med and 1.867\% over LLaDA-V in the open-ended biomedical visual conversation task, and sets new state-of-the-art accuracy on the closed-form subset of three VQA benchmarks: 84.93\% on VQA-RAD, 92.31\% on SLAKE, and 95.15\% on PathVQA. Furthermore, a detailed comparison with LLaVA-Med suggests that LLaDA-MedV is capable of generating reasonably longer responses by explicitly controlling response length, which can lead to more informative outputs. We also conduct an in-depth analysis of both the training and inference stages, highlighting the critical roles of initialization weight selection, fine-tuning strategies, and the interplay between sampling steps and response repetition. The code and model weight is released at https://github.com/LLM-VLM-GSL/LLaDA-MedV.

Yue Deng, Yan Yu, Weiyu Ma et al.

The availability of challenging simulation environments is pivotal for advancing the field of Multi-Agent Reinforcement Learning (MARL). In cooperative MARL settings, the StarCraft Multi-Agent Challenge (SMAC) has gained prominence as a benchmark for algorithms following centralized training with decentralized execution paradigm. However, with continual advancements in SMAC, many algorithms now exhibit near-optimal performance, complicating the evaluation of their true effectiveness. To alleviate this problem, in this work, we highlight a critical issue: the default opponent policy in these environments lacks sufficient diversity, leading MARL algorithms to overfit and exploit unintended vulnerabilities rather than learning robust strategies. To overcome these limitations, we propose SMAC-HARD, a novel benchmark designed to enhance training robustness and evaluation comprehensiveness. SMAC-HARD supports customizable opponent strategies, randomization of adversarial policies, and interfaces for MARL self-play, enabling agents to generalize to varying opponent behaviors and improve model stability. Furthermore, we introduce a black-box testing framework wherein agents are trained without exposure to the edited opponent scripts but are tested against these scripts to evaluate the policy coverage and adaptability of MARL algorithms. We conduct extensive evaluations of widely used and state-of-the-art algorithms on SMAC-HARD, revealing the substantial challenges posed by edited and mixed strategy opponents. Additionally, the black-box strategy tests illustrate the difficulty of transferring learned policies to unseen adversaries. We envision SMAC-HARD as a critical step toward benchmarking the next generation of MARL algorithms, fostering progress in self-play methods for multi-agent systems. Our code is available at https://github.com/devindeng94/smac-hard.

Wenhui Zhu, Peijie Qiu, Xiwen Chen et al.

Multiple Instance Learning (MIL) is a popular weakly-supervised method for various applications, with a particular interest in histological whole slide image (WSI) classification. Due to the gigapixel resolution of WSI, applications of MIL in WSI typically necessitate a two-stage training scheme: first, extract features from the pre-trained backbone and then perform MIL aggregation. However, it is well-known that this suboptimal training scheme suffers from "noisy" feature embeddings from the backbone and inherent weak supervision, hindering MIL from learning rich and generalizable features. However, the most commonly used technique (i.e., dropout) for mitigating this issue has yet to be explored in MIL. In this paper, we empirically explore how effective the dropout can be in MIL. Interestingly, we observe that dropping the top-k most important instances within a bag leads to better performance and generalization even under noise attack. Based on this key observation, we propose a novel MIL-specific dropout method, termed MIL-Dropout, which systematically determines which instances to drop. Experiments on five MIL benchmark datasets and two WSI datasets demonstrate that MIL-Dropout boosts the performance of current MIL methods with a negligible computational cost. The code is available at https://github.com/ChongQingNoSubway/MILDropout.

Yujian Xiong, Mohammad Farazi, Yanxi Chen et al.

Representation learning on large-scale unstructured volumetric and surface meshes poses significant challenges in neuroimaging, especially when models must incorporate diverse vertex-level morphometric descriptors, such as cortical thickness, curvature, sulcal depth, and myelin content, which carry subtle disease-related signals. Current approaches either ignore these clinically informative features or support only a single mesh topology, restricting their use across imaging pipelines. We introduce a hierarchical transformer framework designed for heterogeneous mesh analysis that operates on spatially adaptive tree partitions constructed from simplicial complexes of arbitrary order. This design accommodates both volumetric and surface discretizations within a single architecture, enabling efficient multi-scale attention without topology-specific modifications. A feature projection module maps variable-length per-vertex clinical descriptors into the spatial hierarchy, separating geometric structure from feature dimensionality and allowing seamless integration of different neuroimaging feature sets. Self-supervised pretraining via masked reconstruction of both coordinates and morphometric channels on large unlabeled cohorts yields a transferable encoder backbone applicable to diverse downstream tasks and mesh modalities. We validate our approach on Alzheimer's disease classification and amyloid burden prediction using volumetric brain meshes from ADNI, as well as focal cortical dysplasia detection on cortical surface meshes from the MELD dataset, achieving state-of-the-art results across all benchmarks.

Xuanzhao Dong, Wenhui Zhu, Xiwen Chen et al.

Over the past decade, generative models have demonstrated success in enhancing fundus images. However, the evaluation of these models remains a challenge. A benchmark for fundus image enhancement is needed for three main reasons:(1) Conventional denoising metrics such as PSNR and SSIM fail to capture clinically relevant features, such as lesion preservation and vessel morphology consistency, limiting their applicability in real-world settings; (2) There is a lack of unified evaluation protocols that address both paired and unpaired enhancement methods, particularly those guided by clinical expertise; and (3) An evaluation framework should provide actionable insights to guide future advancements in clinically aligned enhancement models. To address these gaps, we introduce EyeBench-V2, a benchmark designed to bridge the gap between enhancement model performance and clinical utility. Our work offers three key contributions:(1) Multi-dimensional clinical-alignment through downstream evaluations: Beyond standard enhancement metrics, we assess performance across clinically meaningful tasks including vessel segmentation, diabetic retinopathy (DR) grading, generalization to unseen noise patterns, and lesion segmentation. (2) Expert-guided evaluation design: We curate a novel dataset enabling fair comparisons between paired and unpaired enhancement methods, accompanied by a structured manual assessment protocol by medical experts, which evaluates clinically critical aspects such as lesion structure alterations, background color shifts, and the introduction of artificial structures. (3) Actionable insights: Our benchmark provides a rigorous, task-oriented analysis of existing generative models, equipping clinical researchers with the evidence needed to make informed decisions, while also identifying limitations in current methods to inform the design of next-generation enhancement models.

Jingjing Wang, Xiwen Chen, Wenhui Zhu et al.

LLM-powered coding agents spend the majority of their token budget reading repository files, yet much of the retrieved code is irrelevant to the task at hand. Existing learned pruners compress this context with a single-objective sequence labeler, collapsing all facets of code relevance into one score and one transition matrix. We show that this formulation creates a modeling bottleneck: a single CRF transition prior must serve heterogeneous retention patterns, including contiguous semantic spans and sparse structural support lines. We propose LaMR (Latent Multi-Rubric), a structured pruning framework that decomposes code relevance into two interpretable quality dimensions, semantic evidence and dependency support, each modeled by a dedicated CRF with dimension-specific transition dynamics. A mixture-of-experts gating network dynamically weights the per-rubric emissions conditioned on the query, and a final CRF layer on the fused emissions produces the aggregate keep-or-prune decision. To supervise each dimension without additional annotation cost, we derive multi-rubric labels from the existing training corpus via AST-based program analysis, simultaneously denoising the teacher's binary labels. By effectively filtering distracting noise, LaMR frequently matches or even outperforms unpruned full-context baselines. Experiments on four benchmarks (SWE-Bench Verified, SWE-QA, LCC, LongCodeQA) show that LaMR wins 12 of 16 head-to-head multi-turn comparisons. It saves up to 31% more tokens on multi-turn agent tasks and improves Exact Match by up to +3.5 on single-turn tasks, while performance is frequently enhanced by denoising the context, and any remaining drops are marginal.

Xiwen Chen, Wenhui Zhu, Peijie Qiu et al.

Vision-language models (VLMs) such as CLIP demonstrate strong performance but struggle when adapted to downstream tasks. Prompt learning has emerged as an efficient and effective strategy to adapt VLMs while preserving their pre-trained knowledge. However, existing methods still lead to overfitting and degrade zero-shot generalization. To address this challenge, we propose an optimal transport (OT)-guided prompt learning framework that mitigates forgetting by preserving the structural consistency of feature distributions between pre-trained and fine-tuned models. Unlike conventional point-wise constraints, OT naturally captures cross-instance relationships and expands the feasible parameter space for prompt tuning, allowing a better trade-off between adaptation and generalization. Our approach enforces joint constraints on both vision and text representations, ensuring a holistic feature alignment. Extensive experiments on benchmark datasets demonstrate that our simple yet effective method can outperform existing prompt learning strategies in base-to-novel generalization, cross-dataset evaluation, and domain generalization without additional augmentation or ensemble techniques. The code is available at https://github.com/ChongQingNoSubway/Prompt-OT

Xuanzhao Dong, Wenhui Zhu, Xin Li et al.

Retinal fundus photography enhancement is important for diagnosing and monitoring retinal diseases. However, early approaches to retinal image enhancement, such as those based on Generative Adversarial Networks (GANs), often struggle to preserve the complex topological information of blood vessels, resulting in spurious or missing vessel structures. The persistence diagram, which captures topological features based on the persistence of topological structures under different filtrations, provides a promising way to represent the structure information. In this work, we propose a topology-preserving training paradigm that regularizes blood vessel structures by minimizing the differences of persistence diagrams. We call the resulting framework Topology Preserving Optimal Transport (TPOT). Experimental results on a large-scale dataset demonstrate the superiority of the proposed method compared to several state-of-the-art supervised and unsupervised techniques, both in terms of image quality and performance in the downstream blood vessel segmentation task. The code is available at https://github.com/Retinal-Research/TPOT.

Wenhui Zhu, Xuanzhao Dong, Xin Li et al.

Over the past decade, generative models have achieved significant success in enhancement fundus images.However, the evaluation of these models still presents a considerable challenge. A comprehensive evaluation benchmark for fundus image enhancement is indispensable for three main reasons: 1) The existing denoising metrics (e.g., PSNR, SSIM) are hardly to extend to downstream real-world clinical research (e.g., Vessel morphology consistency). 2) There is a lack of comprehensive evaluation for both paired and unpaired enhancement methods, along with the need for expert protocols to accurately assess clinical value. 3) An ideal evaluation system should provide insights to inform future developments of fundus image enhancement. To this end, we propose a novel comprehensive benchmark, EyeBench, to provide insights that align enhancement models with clinical needs, offering a foundation for future work to improve the clinical relevance and applicability of generative models for fundus image enhancement. EyeBench has three appealing properties: 1) multi-dimensional clinical alignment downstream evaluation: In addition to evaluating the enhancement task, we provide several clinically significant downstream tasks for fundus images, including vessel segmentation, DR grading, denoising generalization, and lesion segmentation. 2) Medical expert-guided evaluation design: We introduce a novel dataset that promote comprehensive and fair comparisons between paired and unpaired methods and includes a manual evaluation protocol by medical experts. 3) Valuable insights: Our benchmark study provides a comprehensive and rigorous evaluation of existing methods across different downstream tasks, assisting medical experts in making informed choices. Additionally, we offer further analysis of the challenges faced by existing methods. The code is available at \url{https://github.com/Retinal-Research/EyeBench}

Vamsi Krishna Vasa, Wenhui Zhu, Xiwen Chen et al.

In recent years, significant progress has been made in the medical image analysis domain using convolutional neural networks (CNNs). In particular, deep neural networks based on a U-shaped architecture (UNet) with skip connections have been adopted for several medical imaging tasks, including organ segmentation. Despite their great success, CNNs are not good at learning global or semantic features. Especially ones that require human-like reasoning to understand the context. Many UNet architectures attempted to adjust with the introduction of Transformer-based self-attention mechanisms, and notable gains in performance have been noted. However, the transformers are inherently flawed with redundancy to learn at shallow layers, which often leads to an increase in the computation of attention from the nearby pixels offering limited information. The recently introduced Super Token Attention (STA) mechanism adapts the concept of superpixels from pixel space to token space, using super tokens as compact visual representations. This approach tackles the redundancy by learning efficient global representations in vision transformers, especially for the shallow layers. In this work, we introduce the STA module in the UNet architecture (STA-UNet), to limit redundancy without losing rich information. Experimental results on four publicly available datasets demonstrate the superiority of STA-UNet over existing state-of-the-art architectures in terms of Dice score and IOU for organ segmentation tasks. The code is available at \url{https://github.com/Retinal-Research/STA-UNet}.

Xiwen Chen, Wenhui Zhu, Gen Li et al.

Multi-modal large language models (MLLMs) achieve strong visual-language reasoning but suffer from high inference cost due to redundant visual tokens. Recent work explores visual token pruning to accelerate inference, while existing pruning methods overlook the underlying distributional structure of visual representations. We propose OTPrune, a training-free framework that formulates pruning as distribution alignment via optimal transport (OT). By minimizing the 2-Wasserstein distance between the full and pruned token distributions, OTPrune preserves both local diversity and global representativeness while reducing inference cost. Moreover, we derive a tractable submodular objective that enables efficient optimization, and theoretically prove its monotonicity and submodularity, providing a principled foundation for stable and efficient pruning. We further provide a comprehensive analysis that explains how distributional alignment contributes to stable and semantically faithful pruning. Comprehensive experiments on wider benchmarks demonstrate that OTPrune achieves superior performance-efficiency tradeoffs compared to state-of-the-art methods. The code is available at https://github.com/xiwenc1/OTPrune.

Xuanzhao Dong, Wenhui Zhu, Yujian Xiong et al.

Color fundus photography (CFP) is central to diagnosing and monitoring retinal disease, yet its acquisition variability (e.g., illumination changes) often degrades image quality, which motivates robust enhancement methods. Unpaired enhancement pipelines are typically GAN-based, however, they can distort clinically critical vasculature, altering vessel topology and endpoint integrity. Motivated by these structural alterations, we propose Vessel-Aware Optimal Transport (\textbf{VAOT}), a framework that combines an optimal-transport objective with two structure-preserving regularizers: (i) a skeleton-based loss to maintain global vascular connectivity and (ii) an endpoint-aware loss to stabilize local termini. These constraints guide learning in the unpaired setting, reducing noise while preserving vessel structure. Experimental results on synthetic degradation benchmark and downstream evaluations in vessel and lesion segmentation demonstrate the superiority of the proposed methods against several state-of-the art baselines. The code is available at https://github.com/Retinal-Research/VAOT

Xiwen Chen, Peijie Qiu, Wenhui Zhu et al.

While multiple instance learning (MIL) has shown to be a promising approach for histopathological whole slide image (WSI) analysis, its reliance on permutation invariance significantly limits its capacity to effectively uncover semantic correlations between instances within WSIs. Based on our empirical and theoretical investigations, we argue that approaches that are not permutation-invariant but better capture spatial correlations between instances can offer more effective solutions. In light of these findings, we propose a novel alternative to existing MIL for WSI analysis by learning to restore the order of instances from their randomly shuffled arrangement. We term this task as cracking an instance jigsaw puzzle problem, where semantic correlations between instances are uncovered. To tackle the instance jigsaw puzzles, we propose a novel Siamese network solution, which is theoretically justified by optimal transport theory. We validate the proposed method on WSI classification and survival prediction tasks, where the proposed method outperforms the recent state-of-the-art MIL competitors. The code is available at https://github.com/xiwenc1/MIL-JigsawPuzzles.

Wenhui Zhu, Xiwen Chen, Peijie Qiu et al.

Since its introduction, UNet has been leading a variety of medical image segmentation tasks. Although numerous follow-up studies have also been dedicated to improving the performance of standard UNet, few have conducted in-depth analyses of the underlying interest pattern of UNet in medical image segmentation. In this paper, we explore the patterns learned in a UNet and observe two important factors that potentially affect its performance: (i) irrelative feature learned caused by asymmetric supervision; (ii) feature redundancy in the feature map. To this end, we propose to balance the supervision between encoder and decoder and reduce the redundant information in the UNet. Specifically, we use the feature map that contains the most semantic information (i.e., the last layer of the decoder) to provide additional supervision to other blocks to provide additional supervision and reduce feature redundancy by leveraging feature distillation. The proposed method can be easily integrated into existing UNet architecture in a plug-and-play fashion with negligible computational cost. The experimental results suggest that the proposed method consistently improves the performance of standard UNets on four medical image segmentation datasets. The code is available at \url{https://github.com/ChongQingNoSubway/SelfReg-UNet}

Xiwen Chen, Peijie Qiu, Wenhui Zhu et al.

Deep neural networks, including transformers and convolutional neural networks, have significantly improved multivariate time series classification (MTSC). However, these methods often rely on supervised learning, which does not fully account for the sparsity and locality of patterns in time series data (e.g., diseases-related anomalous points in ECG). To address this challenge, we formally reformulate MTSC as a weakly supervised problem, introducing a novel multiple-instance learning (MIL) framework for better localization of patterns of interest and modeling time dependencies within time series. Our novel approach, TimeMIL, formulates the temporal correlation and ordering within a time-aware MIL pooling, leveraging a tokenized transformer with a specialized learnable wavelet positional token. The proposed method surpassed 26 recent state-of-the-art methods, underscoring the effectiveness of the weakly supervised TimeMIL in MTSC. The code will be available at https://github.com/xiwenc1/TimeMIL.

Xiwen Chen, Wenhui Zhu, Songzhu Zheng et al.

Large language models (LLMs) are increasingly deployed for autonomous financial trading, a domain requiring continuous adaptation to noisy, non-stationary markets. Existing self-improving agents typically address this through unbounded free-form prompt optimization. However, in low signal-to-noise environments with delayed scalar rewards (P\&L), this unstructured approach exacerbates the fundamental credit assignment problem: optimizers cannot reliably distinguish systematic logic flaws from stochastic market variance, inevitably leading to policy drift. To overcome this bottleneck, we introduce the Self-Evolving Human-Auditable Rubric Policy (SHARP), a neuro-symbolic framework that replaces unconstrained text mutation with structured, symbolic policy optimization. SHARP confines the agent's reasoning to a bounded, human-readable rubric of explicit condition-action rules. When sub-optimal trades occur, an attribution agent employs cross-sample reasoning across multiple samples to isolate specific rule failures. This enables targeted, atomic policy edits that are subsequently regularized through strict walk-forward validation. Evaluated across three diverse equity sectors and four LLM backbones, SHARP consistently transforms generic initial heuristics into highly robust strategies, lifting the empirical performance of compact models by 10 to 20 percentage points on average (e.g., GPT-4o-mini). Ultimately, SHARP demonstrates that LLMs can achieve dynamic and efficient adaptation while significantly enhancing the structural transparency and auditability demanded by institutional finance.

Hao Wang, Sayed Pedram Haeri Boroujeni, Xiwen Chen et al.

Wildfires are a significant threat to ecosystems and human infrastructure, leading to widespread destruction and environmental degradation. Recent advancements in deep learning and generative models have enabled new methods for wildfire detection and monitoring. However, the scarcity of annotated wildfire images limits the development of robust models for these tasks. In this work, we present the FLAME Diffuser, a training-free, diffusion-based framework designed to generate realistic wildfire images with paired ground truth. Our framework uses augmented masks, sampled from real wildfire data, and applies Perlin noise to guide the generation of realistic flames. By controlling the placement of these elements within the image, we ensure precise integration while maintaining the original images style. We evaluate the generated images using normalized Frechet Inception Distance, CLIP Score, and a custom CLIP Confidence metric, demonstrating the high quality and realism of the synthesized wildfire images. Specifically, the fusion of Perlin noise in this work significantly improved the quality of synthesized images. The proposed method is particularly valuable for enhancing datasets used in downstream tasks such as wildfire detection and monitoring.

Wenhui Zhu, Xuanzhao Dong, Xin Li et al.

Recently, reinforcement learning (RL)-based tuning has shifted the trajectory of Multimodal Large Language Models (MLLMs), particularly following the introduction of Group Relative Policy Optimization (GRPO). However, directly applying it to medical tasks remains challenging for achieving clinically grounded model behavior. Motivated by the need to align model response with clinical expectations, we investigate four critical dimensions that affect the effectiveness of RL-based tuning in medical visual question answering (VQA): base model initialization strategy, the role of medical semantic alignment, the impact of length-based rewards on long-chain reasoning, and the influence of bias. We conduct extensive experiments to analyze these factors for medical MLLMs, providing new insights into how models are domain-specifically fine-tuned. Additionally, our results also demonstrate that GRPO-based RL tuning consistently outperforms standard supervised fine-tuning (SFT) in both accuracy and reasoning quality.

Wenchao Xu, Jinyu Chen, Peirong Zheng et al.

Foundation model (FM) powered agent services are regarded as a promising solution to develop intelligent and personalized applications for advancing toward Artificial General Intelligence (AGI). To achieve high reliability and scalability in deploying these agent services, it is essential to collaboratively optimize computational and communication resources, thereby ensuring effective resource allocation and seamless service delivery. In pursuit of this vision, this paper proposes a unified framework aimed at providing a comprehensive survey on deploying FM-based agent services across heterogeneous devices, with the emphasis on the integration of model and resource optimization to establish a robust infrastructure for these services. Particularly, this paper begins with exploring various low-level optimization strategies during inference and studies approaches that enhance system scalability, such as parallelism techniques and resource scaling methods. The paper then discusses several prominent FMs and investigates research efforts focused on inference acceleration, including techniques such as model compression and token reduction. Moreover, the paper also investigates critical components for constructing agent services and highlights notable intelligent applications. Finally, the paper presents potential research directions for developing real-time agent services with high Quality of Service (QoS).

Xiwen Chen, Wenhui Zhu, Peijie Qiu et al.

Analyzing time series data is crucial to a wide spectrum of applications, including economics, online marketplaces, and human healthcare. In particular, time series classification plays an indispensable role in segmenting different phases in stock markets, predicting customer behavior, and classifying worker actions and engagement levels. These aspects contribute significantly to the advancement of automated decision-making and system optimization in real-world applications. However, there is a large consensus that time series data often suffers from domain shifts between training and test sets, which dramatically degrades the classification performance. Despite the success of (reversible) instance normalization in handling the domain shifts for time series regression tasks, its performance in classification is unsatisfactory. In this paper, we propose \textit{FIC-TSC}, a training framework for time series classification that leverages Fisher information as the constraint. We theoretically and empirically show this is an efficient and effective solution to guide the model converge toward flatter minima, which enhances its generalizability to distribution shifts. We rigorously evaluate our method on 30 UEA multivariate and 85 UCR univariate datasets. Our empirical results demonstrate the superiority of the proposed method over 14 recent state-of-the-art methods.

Xiwen Chen, Peijie Qiu, Wenhui Zhu et al.

Since its introduction, the transformer has shifted the development trajectory away from traditional models (e.g., RNN, MLP) in time series forecasting, which is attributed to its ability to capture global dependencies within temporal tokens. Follow-up studies have largely involved altering the tokenization and self-attention modules to better adapt Transformers for addressing special challenges like non-stationarity, channel-wise dependency, and variable correlation in time series. However, we found that the expressive capability of sequence representation is a key factor influencing Transformer performance in time forecasting after investigating several representative methods, where there is an almost linear relationship between sequence representation entropy and mean square error, with more diverse representations performing better. In this paper, we propose a novel attention mechanism with Sequence Complementors and prove feasible from an information theory perspective, where these learnable sequences are able to provide complementary information beyond current input to feed attention. We further enhance the Sequence Complementors via a diversification loss that is theoretically covered. The empirical evaluation of both long-term and short-term forecasting has confirmed its superiority over the recent state-of-the-art methods.

Peijie Qiu, Wenhui Zhu, Sayantan Kumar et al.

Multiple signal modalities, such as vision and sounds, are naturally present in real-world phenomena. Recently, there has been growing interest in learning generative models, in particular variational autoencoder (VAE), to for multimodal representation learning especially in the case of missing modalities. The primary goal of these models is to learn a modality-invariant and modality-specific representation that characterizes information across multiple modalities. Previous attempts at multimodal VAEs approach this mainly through the lens of experts, aggregating unimodal inference distributions with a product of experts (PoE), a mixture of experts (MoE), or a combination of both. In this paper, we provide an alternative generic and theoretical formulation of multimodal VAE through the lens of barycenter. We first show that PoE and MoE are specific instances of barycenters, derived by minimizing the asymmetric weighted KL divergence to unimodal inference distributions. Our novel formulation extends these two barycenters to a more flexible choice by considering different types of divergences. In particular, we explore the Wasserstein barycenter defined by the 2-Wasserstein distance, which better preserves the geometry of unimodal distributions by capturing both modality-specific and modality-invariant representations compared to KL divergence. Empirical studies on three multimodal benchmarks demonstrated the effectiveness of the proposed method.