Xingjian Diao

Xingjian Diao, Ming Cheng, Shitong Cheng

Learning high-quality video representation has shown significant applications in computer vision and remains challenging. Previous work based on mask autoencoders such as ImageMAE and VideoMAE has proven the effectiveness of learning representations in images and videos through reconstruction strategy in the visual modality. However, these models exhibit inherent limitations, particularly in scenarios where extracting features solely from the visual modality proves challenging, such as when dealing with low-resolution and blurry original videos. Based on this, we propose AV-MaskEnhancer for learning high-quality video representation by combining visual and audio information. Our approach addresses the challenge by demonstrating the complementary nature of audio and video features in cross-modality content. Moreover, our result of the video classification task on the UCF101 dataset outperforms the existing work and reaches the state-of-the-art, with a top-1 accuracy of 98.8% and a top-5 accuracy of 99.9%.

Xingjian Diao, Chunhui Zhang, Weiyi Wu et al.

Multimodal foundation models (MFMs) have demonstrated significant success in tasks such as visual captioning, question answering, and image-text retrieval. However, these models face inherent limitations due to their finite internal capacity, which restricts their ability to process extended temporal sequences, a crucial requirement for comprehensive video and audio analysis. To overcome these challenges, we introduce a specialized cognitive module, temporal working memory (TWM), which aims to enhance the temporal modeling capabilities of MFMs. It selectively retains task-relevant information across temporal dimensions, ensuring that critical details are preserved throughout the processing of video and audio content. The TWM uses a query-guided attention approach to focus on the most informative multimodal segments within temporal sequences. By retaining only the most relevant content, TWM optimizes the use of the model's limited capacity, enhancing its temporal modeling ability. This plug-and-play module can be easily integrated into existing MFMs. With our TWM, nine state-of-the-art models exhibit significant performance improvements across tasks such as video captioning, question answering, and video-text retrieval. By enhancing temporal modeling, TWM extends the capability of MFMs to handle complex, time-sensitive data effectively. Our code is available at https://github.com/xid32/NAACL_2025_TWM.

Xingjian Diao, Chunhui Zhang, Tingxuan Wu et al.

Music performances are representative scenarios for audio-visual modeling. Unlike common scenarios with sparse audio, music performances continuously involve dense audio signals throughout. While existing multimodal learning methods on the audio-video QA demonstrate impressive capabilities in general scenarios, they are incapable of dealing with fundamental problems within the music performances: they underexplore the interaction between the multimodal signals in performance and fail to consider the distinctive characteristics of instruments and music. Therefore, existing methods tend to answer questions regarding musical performances inaccurately. To bridge the above research gaps, (i) given the intricate multimodal interconnectivity inherent to music data, our primary backbone is designed to incorporate multimodal interactions within the context of music; (ii) to enable the model to learn music characteristics, we annotate and release rhythmic and music sources in the current music datasets; (iii) for time-aware audio-visual modeling, we align the model's music predictions with the temporal dimension. Our experiments show state-of-the-art effects on the Music AVQA datasets. Our code is available at https://github.com/xid32/Amuse.

Peijun Qing, Chongyang Gao, Yefan Zhou et al.

Parameter-efficient fine-tuning methods, such as Low-Rank Adaptation (LoRA), are known to enhance training efficiency in Large Language Models (LLMs). Due to the limited parameters of LoRA, recent studies seek to combine LoRA with Mixture-of-Experts (MoE) to boost performance across various tasks. However, inspired by the observed redundancy in traditional MoE structures, previous studies identify similar redundancy among LoRA experts within the MoE architecture, highlighting the necessity for non-uniform allocation of LoRA experts across different layers. In this paper, we leverage Heavy-Tailed Self-Regularization (HT-SR) Theory to design a fine-grained allocation strategy. Our analysis reveals that the number of experts per layer correlates with layer training quality, which exhibits significant variability across layers. Based on this, we introduce AlphaLoRA, a theoretically principled and training-free method for allocating LoRA experts to further mitigate redundancy. Experiments on three models across ten language processing and reasoning benchmarks demonstrate that AlphaLoRA achieves comparable or superior performance over all baselines. Our code is available at https://github.com/morelife2017/alphalora.

Xingjian Diao, Chunhui Zhang, Keyi Kong et al.

While large language models have demonstrated impressive reasoning abilities, their extension to the audio modality, particularly within large audio-language models (LALMs), remains underexplored. Addressing this gap requires a systematic approach that involves a capable base model, high-quality reasoning-oriented audio data, and effective training algorithms. In this work, we present a comprehensive solution for audio logical reasoning (ALR) tasks: we introduce SoundMind, a dataset of 6,446 audio-text annotated samples specifically curated to support complex reasoning. Building on this resource, we propose SoundMind-RL, a rule-based reinforcement learning (RL) algorithm designed to equip audio-language models with robust audio-text reasoning capabilities. By fine-tuning Qwen2.5-Omni-7B on the proposed SoundMind dataset using SoundMind-RL, we achieve strong and consistent improvements over state-of-the-art baselines on the SoundMind benchmark. This work highlights the benefit of combining high-quality, reasoning-focused datasets with specialized RL techniques, and contributes to advancing auditory intelligence in language models. The code and dataset introduced in this work are publicly available at https://github.com/xid32/SoundMind.

Wenhao You, Xingjian Diao, Chunhui Zhang et al.

While recent Multimodal Large Language Models exhibit impressive capabilities for general multimodal tasks, specialized domains like music necessitate tailored approaches. Music Audio-Visual Question Answering (Music AVQA) particularly underscores this, presenting unique challenges with its continuous, densely layered audio-visual content, intricate temporal dynamics, and the critical need for domain-specific knowledge. Through a systematic analysis of Music AVQA datasets and methods, this position paper identifies that specialized input processing, architectures incorporating dedicated spatial-temporal designs, and music-specific modeling strategies are critical for success in this domain. Our study provides valuable insights for researchers by highlighting effective design patterns empirically linked to strong performance, proposing concrete future directions for incorporating musical priors, and aiming to establish a robust foundation for advancing multimodal musical understanding. This work is intended to inspire broader attention and further research, supported by a continuously updated anonymous GitHub repository of relevant papers: https://github.com/xid32/Survey4MusicAVQA.

Weiyi Wu, Xinwen Xu, Chongyang Gao et al.

Whole slide images, with their gigapixel-scale panoramas of tissue samples, are pivotal for precise disease diagnosis. However, their analysis is hindered by immense data size and scarce annotations. Existing MIL methods face challenges due to the fundamental imbalance where a single bag-level label must guide the learning of numerous patch-level features. This sparse supervision makes it difficult to reliably identify discriminative patches during training, leading to unstable optimization and suboptimal solutions. We propose a spatially regularized MIL framework that leverages inherent spatial relationships among patch features as label-independent regularization signals. Our approach learns a shared representation space by jointly optimizing feature-induced spatial reconstruction and label-guided classification objectives, enforcing consistency between intrinsic structural patterns and supervisory signals. Experimental results on multiple public datasets demonstrate significant improvements over state-of-the-art methods, offering a promising direction.

Weiyi Wu, Xinwen Xu, Chongyang Gao et al.

Large Language Models (LLMs) have great potential in the field of health care, yet they face great challenges in adapting to rapidly evolving medical knowledge. This can lead to outdated or contradictory treatment suggestions. This study investigated how LLMs respond to evolving clinical guidelines, focusing on concept drift and internal inconsistencies. We developed the DriftMedQA benchmark to simulate guideline evolution and assessed the temporal reliability of various LLMs. Our evaluation of seven state-of-the-art models across 4,290 scenarios demonstrated difficulties in rejecting outdated recommendations and frequently endorsing conflicting guidance. Additionally, we explored two mitigation strategies: Retrieval-Augmented Generation and preference fine-tuning via Direct Preference Optimization. While each method improved model performance, their combination led to the most consistent and reliable results. These findings underscore the need to improve LLM robustness to temporal shifts to ensure more dependable applications in clinical practice. The dataset is available at https://huggingface.co/datasets/RDBH/DriftMed.

Xingjian Diao, Ming Cheng, Wayner Barrios et al.

Talking face generation has gained immense popularity in the computer vision community, with various applications including AR, VR, teleconferencing, digital assistants, and avatars. Traditional methods are mainly audio-driven, which have to deal with the inevitable resource-intensive nature of audio storage and processing. To address such a challenge, we propose FT2TF - First-Person Statement Text-To-Talking Face Generation, a novel one-stage end-to-end pipeline for talking face generation driven by first-person statement text. Different from previous work, our model only leverages visual and textual information without any other sources (e.g., audio/landmark/pose) during inference. Extensive experiments are conducted on LRS2 and LRS3 datasets, and results on multi-dimensional evaluation metrics are reported. Both quantitative and qualitative results showcase that FT2TF outperforms existing relevant methods and reaches the state-of-the-art. This achievement highlights our model's capability to bridge first-person statements and dynamic face generation, providing insightful guidance for future work.

Ming Cheng, Ziyi Zhou, Bowen Zhang et al.

In the landscape of spatio-temporal data analytics, effective trajectory representation learning is paramount. To bridge the gap of learning accurate representations with efficient and flexible mechanisms, we introduce Efflex, a comprehensive pipeline for transformative graph modeling and representation learning of the large-volume spatio-temporal trajectories. Efflex pioneers the incorporation of a multi-scale k-nearest neighbors (KNN) algorithm with feature fusion for graph construction, marking a leap in dimensionality reduction techniques by preserving essential data features. Moreover, the groundbreaking graph construction mechanism and the high-performance lightweight GCN increase embedding extraction speed by up to 36 times faster. We further offer Efflex in two versions, Efflex-L for scenarios demanding high accuracy, and Efflex-B for environments requiring swift data processing. Comprehensive experimentation with the Porto and Geolife datasets validates our approach, positioning Efflex as the state-of-the-art in the domain. Such enhancements in speed and accuracy highlight the versatility of Efflex, underscoring its wide-ranging potential for deployment in time-sensitive and computationally constrained applications.

Xingjian Diao, Tianzhen Yang, Chunhui Zhang et al.

Music performances, characterized by dense and continuous audio as well as seamless audio-visual integration, present unique challenges for multimodal scene understanding and reasoning. Recent Music Performance Audio-Visual Question Answering (Music AVQA) datasets have been proposed to reflect these challenges, highlighting the continued need for more effective integration of audio-visual representations in complex question answering. However, existing Music AVQA methods often rely on dense and unoptimized representations, leading to inefficiencies in the isolation of key information, the reduction of redundancy, and the prioritization of critical samples. To address these challenges, we introduce Sparsify, a sparse learning framework specifically designed for Music AVQA. It integrates three sparsification strategies into an end-to-end pipeline and achieves state-of-the-art performance on the Music AVQA datasets. In addition, it reduces training time by 28.32% compared to its fully trained dense counterpart while maintaining accuracy, demonstrating clear efficiency gains. To further improve data efficiency, we propose a key-subset selection algorithm that selects and uses approximately 25% of MUSIC-AVQA v2.0 training data and retains 70-80% of full-data performance across models.

Ziyi Zhou, Ming Cheng, Xingjian Diao et al.

The escalating prevalence of diabetes globally underscores the need for diabetes management. Recent research highlights the growing focus on digital biomarkers in diabetes management, with innovations in computational frameworks and noninvasive monitoring techniques using personalized glucose metrics. However, they predominantly focus on insulin dosing and specific glucose values, or with limited attention given to overall glycemic control. This leaves a gap in expanding the scope of digital biomarkers for overall glycemic control in diabetes management. To address such a research gap, we propose GluMarker -- an end-to-end framework for modeling digital biomarkers using broader factors sources to predict glycemic control. Through the assessment and refinement of various machine learning baselines, GluMarker achieves state-of-the-art on Anderson's dataset in predicting next-day glycemic control. Moreover, our research identifies key digital biomarkers for the next day's glycemic control prediction. These identified biomarkers are instrumental in illuminating the daily factors that influence glycemic management, offering vital insights for diabetes care.

Ming Cheng, Xingjian Diao, Ziyi Zhou et al.

The global diabetes epidemic highlights the importance of maintaining good glycemic control. Glucose prediction is a fundamental aspect of diabetes management, facilitating real-time decision-making. Recent research has introduced models focusing on long-term glucose trend prediction, which are unsuitable for real-time decision-making and result in delayed responses. Conversely, models designed to respond to immediate glucose level changes cannot analyze glucose variability comprehensively. Moreover, contemporary research generally integrates various physiological parameters (e.g. insulin doses, food intake, etc.), which inevitably raises data privacy concerns. To bridge such a research gap, we propose TimeGlu -- an end-to-end pipeline for short-term glucose prediction solely based on CGM time series data. We implement four baseline methods to conduct a comprehensive comparative analysis of the model's performance. Through extensive experiments on two contrasting datasets (CGM Glucose and Colas dataset), TimeGlu achieves state-of-the-art performance without the need for additional personal data from patients, providing effective guidance for real-world diabetic glucose management.

Ziyi Zhou, Ming Cheng, Yanjun Cui et al.

The increasing number of diabetic patients is a serious issue in society today, which has significant negative impacts on people's health and the country's financial expenditures. Because diabetes may develop into potential serious complications, early glucose prediction for diabetic patients is necessary for timely medical treatment. Existing glucose prediction methods typically utilize patients' private data (e.g. age, gender, ethnicity) and physiological parameters (e.g. blood pressure, heart rate) as reference features for glucose prediction, which inevitably leads to privacy protection concerns. Moreover, these models generally focus on either long-term (monthly-based) or short-term (minute-based) predictions. Long-term prediction methods are generally inaccurate because of the external uncertainties that can greatly affect the glucose values, while short-term ones fail to provide timely medical guidance. Based on the above issues, we propose CrossGP, a novel machine-learning framework for cross-day glucose prediction solely based on the patient's external activities without involving any physiological parameters. Meanwhile, we implement three baseline models for comparison. Extensive experiments on Anderson's dataset strongly demonstrate the superior performance of CrossGP and prove its potential for future real-life applications.

Ming Cheng, Bowen Zhang, Ziyu Wang et al.

Trajectory similarity search plays an essential role in autonomous driving, as it enables vehicles to analyze the information and characteristics of different trajectories to make informed decisions and navigate safely in dynamic environments. Existing work on the trajectory similarity search task primarily utilizes sequence-processing algorithms or Recurrent Neural Networks (RNNs), which suffer from the inevitable issues of complicated architecture and heavy training costs. Considering the intricate connections between trajectories, using Graph Neural Networks (GNNs) for data modeling is feasible. However, most methods directly use existing mathematical graph structures as the input instead of constructing specific graphs from certain vehicle trajectory data. This ignores such data's unique and dynamic characteristics. To bridge such a research gap, we propose VeTraSS -- an end-to-end pipeline for Vehicle Trajectory Similarity Search. Specifically, VeTraSS models the original trajectory data into multi-scale graphs, and generates comprehensive embeddings through a novel multi-layer attention-based GNN. The learned embeddings can be used for searching similar vehicle trajectories. Extensive experiments on the Porto and Geolife datasets demonstrate the effectiveness of VeTraSS, where our model outperforms existing work and reaches the state-of-the-art. This demonstrates the potential of VeTraSS for trajectory analysis and safe navigation in self-driving vehicles in the real world.

Xingjian Diao, Zheyuan Liu, Chunhui Zhang et al.

Large Vision-Language Models (LVLMs) have exhibited strong reasoning capabilities through chain-of-thought mechanisms that generate step-by-step rationales. However, such slow-thinking approaches often lead to overthinking, where models produce excessively verbose responses even for simple queries, resulting in test-time inefficiency and even degraded accuracy. Prior work has attempted to mitigate this issue via adaptive reasoning strategies, but these methods largely overlook a fundamental bottleneck: visual perception failures. We argue that stable reasoning critically depends on low-level visual grounding, and that reasoning errors often originate from imperfect perception rather than insufficient deliberation. To address this limitation, we propose Gated Perception-Reasoning Optimization (GPRO), a meta-reasoning controller that dynamically routes computation among three decision paths at each generation step: a lightweight fast path, a slow perception path for re-examining visual inputs, and a slow reasoning path for internal self-reflection. To learn this distinction, we derive large-scale failure attribution supervision from approximately 790k samples, using teacher models to distinguish perceptual hallucinations from reasoning errors. We then train the controller with multi-objective reinforcement learning to optimize the trade-off between task accuracy and computational cost under uncertainty. Experiments on five benchmarks demonstrate that GPRO substantially improves both accuracy and efficiency, outperforming recent slow-thinking methods while generating significantly shorter responses.

Yaning Jia, Chunhui Zhang, Xingjian Diao et al.

Curriculum learning (CL) - ordering training data from easy to hard - has become a popular strategy for improving reasoning in large language models (LLMs). Yet prior work employs disparate difficulty metrics and training setups, leaving open fundamental questions: When does curriculum help? Which direction - forward or reverse - is better? And does the answer depend on what we measure? We address these questions through a unified offline evaluation framework that decomposes curriculum difficulty into five complementary dimensions: Problem Difficulty, Model Surprisal, Confidence Margin, Predictive Uncertainty, and Decision Variability. Through controlled post-training experiments on mathematical reasoning benchmarks with Llama3.1-8B, Mistral-7B, and Gemma3-4B, we find that (i) no curriculum strategy dominates universally - the relative effectiveness of forward versus reverse CL depends jointly on model capability and task complexity; (ii) even within a single metric, samples at different difficulty levels produce distinct gains depending on task demands; and (iii) task-aligned curricula focus on shaping the model's final representations and generalization, whereas inner-state curricula modulate internal states such as confidence and uncertainty. Our findings challenge the notion of a universal curriculum strategy and offer actionable guidance across model and task regimes, with some metrics indicating that prioritizing decision-uncertain samples can further enhance learning outcomes.

Xingjian Diao, Weiyi Wu, Keyi Kong et al.

Visual Question Answering (VQA) is increasingly used in diverse applications ranging from general visual reasoning to safety-critical domains such as medical imaging and autonomous systems, where models must provide not only accurate answers but also explanations that humans can easily understand and verify. Prototype-based modeling has shown promise for interpretability by grounding predictions in semantically meaningful regions for purely visual reasoning tasks, yet remains underexplored in the context of VQA. We present ProtoVQA, a unified prototypical framework that (i) learns question-aware prototypes that serve as reasoning anchors, connecting answers to discriminative image regions, (ii) applies spatially constrained matching to ensure that the selected evidence is coherent and semantically relevant, and (iii) supports both answering and grounding tasks through a shared prototype backbone. To assess explanation quality, we propose the Visual-Linguistic Alignment Score (VLAS), which measures how well the model's attended regions align with ground-truth evidence. Experiments on Visual7W show that ProtoVQA yields faithful, fine-grained explanations while maintaining competitive accuracy, advancing the development of transparent and trustworthy VQA systems.

Weiyi Wu, Xingjian Diao, Chongyang Gao et al.

Whole slide images (WSIs) present fundamental computational challenges due to their gigapixel-scale resolutions and sparse, irregularly distributed informative regions. Conventional patch-based methods inevitably distort spatial relationships or treat patches as independent samples, while traditional attention mechanisms, designed for dense, uniformly distributed data, are computationally impractical for WSIs. To address these limitations, we propose a novel sparse-native computational framework that preserves exact spatial relationships, unlocking advanced modeling techniques and bridging a long-standing gap between WSI analysis and general vision. Based on this framework, we develop Sparse Pyramid Attention Networks (SPAN), incorporating a hierarchical sparse pyramid attention architecture with shifted windows that efficiently directs computational resources to informative regions. SPAN comprises two key modules: Spatial-Adaptive Feature Condensation, which progressively builds multi-scale representations from a single-scale input through sparse downsampling, and Context-Aware Feature Refinement, which captures long-range dependencies via shifted windows and global tokens. Evaluations on multiple public datasets demonstrate SPAN's superior performance over state-of-the-art methods, validating both our framework's effectiveness and SPAN's specific advantages in capturing contextual and hierachical representations that existing methods fundamentally cannot model. Our work establishes a new paradigm for WSI analysis that overcomes long-standing computational barriers. The code will be made publicly available upon publication.

Lin Shi, Chiyu Ma, Wenhua Liang et al.

LLM-as-a-Judge has emerged as a promising alternative to human evaluators across various tasks, yet inherent biases - particularly position bias, the tendency to favor solutions based on their position within the prompt - compromise its reliability. This exploratory study evaluates position bias in LLM judges across pairwise and list-wise comparison settings, introducing three metrics: repetition stability, position consistency, and preference fairness. Our experiments, involving 15 LLM judges across MTBench and DevBench with 22 tasks and approximately 40 solution-generating models, result in over 150,000 evaluation instances. We identify Judge-Level, Candidate-Level, and Task-Level factors contributing to bias. The findings confirm that position bias is not due to random chance and varies significantly across judges and tasks. While position bias is weakly influenced by the length of prompt components, it is strongly affected by the quality gap between solutions. Our agreement and disagreement analysis among judges further provides insights into the distribution of judging difficulty across the dataset, and highlights the potential for dataset modifications.

Ming Cheng, Xingjian Diao, Shitong Cheng et al.

Speech anonymization and de-identification have garnered significant attention recently, especially in the healthcare area including telehealth consultations, patient voiceprint matching, and patient real-time monitoring. Speaker identity classification tasks, which involve recognizing specific speakers from audio to learn identity features, are crucial for de-identification. Since rare studies have effectively combined speech anonymization with identity classification, we propose SAIC - an innovative pipeline for integrating Speech Anonymization and Identity Classification. SAIC demonstrates remarkable performance and reaches state-of-the-art in the speaker identity classification task on the Voxceleb1 dataset, with a top-1 accuracy of 96.1%. Although SAIC is not trained or evaluated specifically on clinical data, the result strongly proves the model's effectiveness and the possibility to generalize into the healthcare area, providing insightful guidance for future work.