Author name cluster

Yu Pan

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

18 papers

2 author rows

JBHI Journal 2026 Journal Article

STAND-Net: A Spiking Temporal Attention autoeNcoDer Network for Efficient EEG Artifact Removal

Ruizhi Zhang
Xiaoyu Guo
Yu Pan
Shuo Gao

Electroencephalography (EEG)-based brain computer interface (BCI) systems hold significant promise across diverse applications; however, their performance is compromised by pervasive physiological artifacts that degrade signal fidelity. While current deep neural networks (DNNs) improve artifact rejection, their high computational cost precludes deployment in wearable BCIs systems. Here, we introduce STAND-Net (Spiking Temporal Attention autoeNcoDer Network), a neuromorphic architecture that leverages event-driven spiking neurons to achieve ultra-efficient, high-fidelity EEG artifact removal. STAND-Net combines a spike-convolution encoder-decoder with leaky integrate-and-fire neurons to model spatiotemporal EEG dynamics, a dilation-enhanced residual backbone capturing long-range dependencies, and a spike-rate attention mechanism dynamically localizing artifacts via neuronal firing patterns. The system demonstrates >3. 7 dB improvement in signal-to-distortion ratio over state-of-the-art methods across diverse artifacts while consuming 97. 98% less power than comparable DNNs. Crucially, downstream BCI classification accuracy increased by 6. 64% using STAND-Net-processed signals. This work establishes a neuromorphic framework for low-power and high quality EEG artifact removal in wearable BCI systems.

Details DOI

TMLR Journal 2025 Journal Article

Enhanced Federated Optimization: Adaptive Unbiased Client Sampling with Reduced Variance

Dun Zeng
Zenglin Xu
Yu Pan
Xu Luo
Qifan Wang
Xiaoying Tang

Federated Learning (FL) is a distributed learning paradigm to train a global model across multiple devices without collecting local data. In FL, a server typically selects a subset of clients for each training round to optimize resource usage. Central to this process is the technique of unbiased client sampling, which ensures a representative selection of clients. Current methods primarily utilize a random sampling procedure which, despite its effectiveness, achieves suboptimal efficiency owing to the loose upper bound caused by the sampling variance. In this work, by adopting an independent sampling procedure, we propose a federated optimization framework focused on adaptive unbiased client sampling, improving the convergence rate via an online variance reduction strategy. In particular, we present the first adaptive client sampler, K-Vib, employing an independent sampling procedure. K-Vib achieves a linear speed-up on the regret bound $\tilde{\mathcal{O}}\big(N^{\frac{1}{3}}T^{\frac{2}{3}}/K^{\frac{4}{3}}\big)$ within a set communication budget $K$. Empirical studies indicate that K-Vib doubles the speed compared to baseline algorithms, demonstrating significant potential in federated optimization.

PDF Details

NeurIPS Conference 2025 Conference Paper

Global Prompt Refinement with Non-Interfering Attention Masking for One-Shot Federated Learning

Zhuang Qi
Yu Pan
Lei Meng
Sijin Zhou
Han Yu
Xiaoxiao Li
Xiangxu Meng

Federated Prompt Learning (FPL) enables communication-efficient adaptation by tuning lightweight prompts on top of frozen pre-trained models. Existing FPL methods typically rely on global information, which is only available after the second training round, to facilitate collaboration among client models. Therefore, they are inherently dependent on multi-round communication to fully exhibit their strengths. Moreover, existing one-shot federated learning methods typically focus on fitting seen tasks, but lack cross-task generalization. To bridge this gap, we propose the global prompt refinement with non-interfering attention masking (GPR-NIAM) method for one-shot FPL. The core idea is to design a masking mechanism that restricts excessive interaction between the original text embeddings and the learnable prompt embeddings. GPR-NIAM achieves this through the collaboration of two key modules. Firstly, the attention isolation module suppresses attention from the learnable prompt tokens to the original text tokens, and reweights the reverse attention which preserves generalization across tasks. Secondly, the cross-silo collaborative refinement module integrates decentralized visual knowledge into a unified base and calibrates the global prompt through multi-source cross-modal knowledge alignment, further mitigating the inconsistency caused by data heterogeneity. Extensive experiments conducted on ten benchmark datasets under two tasks show that GPR-NIAM outperforms eight state-of-the-art methods in both class-level and domain-level generalization.

PDF Details

ICML Conference 2025 Conference Paper

Reward Modeling with Ordinal Feedback: Wisdom of the Crowd

Shang Liu
Yu Pan
Guanting Chen 0001
Xiaocheng Li

The canonical setup of learning a reward model (RM) from human preferences with binary feedback discards potentially useful samples (such as "tied" between the two responses) and loses fine-grained information (such as "slightly better’"). This paper proposes a framework for learning RMs under ordinal feedback, generalizing the binary feedback to arbitrary granularity. We first identify a marginal unbiasedness condition, which generalizes the existing assumption of the binary feedback. The condition is validated via the sociological concept called "wisdom of the crowd". Under this condition, we develop a natural probability model and prove the benefits of fine-grained feedback in terms of reducing the Rademacher complexity, which may be of independent interest to another problem: the bias-variance trade-off in knowledge distillation. The framework also sheds light on designing guidelines for human annotators. Our numerical experiments validate that: (1) fine-grained feedback leads to better RM learning for both in- and out-of-distribution settings; (2) incorporating a certain proportion of tied samples boosts RM learning.

Details

AAAI Conference 2025 Conference Paper

StableVC: Style Controllable Zero-Shot Voice Conversion with Conditional Flow Matching

Jixun Yao
Yang Yuguang
Yu Pan
Ziqian Ning
Jianhao Ye
Hongbin Zhou
Lei Xie

Zero-shot voice conversion (VC) aims to transfer the timbre from the source speaker to an arbitrary unseen speaker while preserving the original linguistic content. Despite recent advancements in zero-shot VC using language model-based or diffusion-based approaches, several challenges remain: 1) current approaches primarily focus on adapting timbre from unseen speakers and are unable to transfer style and timbre to different unseen speakers independently; 2) these approaches often suffer from slower inference speeds due to the autoregressive modeling methods or the need for numerous sampling steps; 3) the quality and similarity of the converted samples are still not fully satisfactory. To address these challenges, we propose a Style controllable zero-shot VC approach named StableVC, which aims to transfer timbre and style from source speech to different unseen target speakers. Specifically, we decompose speech into linguistic content, timbre, and style, and then employ a conditional flow matching module to reconstruct the high-quality mel-spectrogram based on these decomposed features. To effectively capture timbre and style in a zero-shot manner, we introduce a novel dual attention mechanism with an adaptive gate, rather than using conventional feature concatenation. With this non-autoregressive design, StableVC can efficiently capture the intricate timbre and style from different unseen speakers and generate high-quality speech significantly faster than real-time. Experiments demonstrate that our proposed StableVC outperforms state-of-the-art baseline systems in zero-shot VC and achieves flexible control over timbre and style from different unseen speakers. Moreover, StableVC offers approximately 25x and 1.65x faster sampling compared to autoregressive and diffusion-based baselines.

PDF Details DOI

NeurIPS Conference 2025 Conference Paper

What Matters in Data for DPO?

Yu Pan
Zhongze Cai
Huaiyang Zhong
Guanting Chen
Chonghuan Wang

Direct Preference Optimization (DPO) has emerged as a simple and effective approach for aligning large language models (LLMs) with human preferences, bypassing the need for a learned reward model. Despite its growing adoption, a fundamental question remains open: what characteristics of preference data are most critical for DPO performance? In this work, we provide a systematic study of how preference data distribution influences DPO, from both theoretical and empirical perspectives. We show that the quality of chosen responses plays a dominant role in optimizing the DPO objective, while the quality of rejected responses may have relatively limited impact. Our theoretical analysis characterizes the optimal response distribution under DPO and reveals how contrastiveness between responses helps primarily by improving the chosen samples. We further study an online DPO setting and show it effectively reduces to supervised fine-tuning on the chosen responses. Extensive experiments across diverse tasks confirm our findings: improving the quality of chosen responses consistently boosts performance regardless of the quality of the rejected responses. We also investigate the benefit of mixing the on-policy data. Our results interpret the mechanism behind some widely adopted strategies and offer practical insights for constructing high-impact preference datasets for LLM alignment.

PDF Details

ICRA Conference 2024 Conference Paper

Active Collision-Based Navigation for Wheeled Robots

Jingjing Li
Jialin Ji
Qianhao Wang
Huan Yu 0002
Yu Pan
Fei Gao 0011

Collision is typically avoided in robot navigation for safety guarantee. However, when a robot’s exteroceptive sensors fail, which means it becomes "blind", collision can actually be leveraged to improve localization performance. Our research demonstrates the informative nature of collisions in this context. Moreover, we show that a robot is able to navigate in a known environment with only proprioceptive sensors by actively colliding with its surroundings for more reliable localization. Firstly, we design a collision-based observation model, which is differentiable and can be easily applied to various estimators. Secondly, we integrate this model into a collision-aided localization framework and implement it in two widely used estimators, the Kalman filter and the particle filter. Thirdly, we propose an active collision path planning method, which effectively reduces localization uncertainty.

Details

AAAI Conference 2024 Conference Paper

Preparing Lessons for Progressive Training on Language Models

Yu Pan
Ye Yuan
Yichun Yin
Jiaxin Shi
Zenglin Xu
Ming Zhang
Lifeng Shang
Xin Jiang

The rapid progress of Transformers in artificial intelligence has come at the cost of increased resource consumption and greenhouse gas emissions due to growing model sizes. Prior work suggests using pretrained small models to improve training efficiency, but this approach may not be suitable for new model structures. On the other hand, training from scratch can be slow, and progressively stacking layers often fails to achieve significant acceleration. To address these challenges, we propose a novel method called Apollo, which prepares lessons for expanding operations by learning high-layer functionality during training of low layers. Our approach involves low-value-prioritized sampling (LVPS) to train different depths and weight sharing to facilitate efficient expansion. We also introduce an interpolation method for stable model depth extension. Experiments demonstrate that Apollo achieves state-of-the-art acceleration ratios, even rivaling methods using pretrained models, making it a universal and efficient solution for training deep models while reducing time, financial, and environmental costs.

PDF Details DOI

JBHI Journal 2024 Journal Article

Temporal Dynamics and Physical Priori Multimodal Network for Rehabilitation Physical Training Evaluation

Shuo Gao
Jianan Chen
Xuhang Chen
Julie Uchitel
Chenyu Tang
Cong Li
Yu Pan
Hubin Zhao

Sensor-based rehabilitation physical training assessment methods have attracted significant attention in refined evaluation scenarios. A refined rehabilitation evaluation method combines the expertise of clinicians with advanced sensor-based technology to capture and analyze subtle movement variations often unobserved by traditional subjective methods. Current approaches center on either body postures or muscle strength, which lack more sophisticated analysis features of muscle activation and coordination, thereby hindering analysis efficacy in deep rehabilitation feature exploration. To address this issue, we present a multimodal network algorithm that integrates surface electromyography (sEMG) and stress distribution signals. The algorithm considers the physical knowledge a priori to interpret the current rehabilitation stage and efficiently handles temporal dynamics arising from diverse user profiles in an online setting. Besides, we verified the performance of this model using a learned-nonuse phenomenon assessment task in 24 subjects, achieving an accuracy of 94. 7%. Our results surpass those of conventional feature-based, distance-based, and ensemble baseline models, highlighting the advantages of incorporating multimodal information rather than relying solely on unimodal data. Moreover, the proposed model presents a network design solution for rehabilitation physical training that requires deep bioinformatic features and can potentially assist real-time and home-based physical training work.

Details DOI

YNIMG Journal 2023 Journal Article

CAMBA framework: Unveiling the brain asymmetry alterations and longitudinal changes after stroke using resting-state EEG

Zexuan Hao
Xiaoxue Zhai
Bo Peng
Dandan Cheng
Yanlin Zhang
Yu Pan
Weibei Dou

Hemispheric asymmetry or lateralization is a fundamental principle of brain organization. However, it is poorly understood to what extent the brain asymmetries across different levels of functional organizations are evident in health or altered in brain diseases. Here, we propose a framework that integrates three degrees of brain interactions (isolated nodes, node-node, and edge-edge) into a unified analysis pipeline to capture the sliding window-based asymmetry dynamics at both the node and hemisphere levels. We apply this framework to resting-state EEG in healthy and stroke populations and investigate the stroke-induced abnormal alterations in brain asymmetries and longitudinal asymmetry changes during poststroke rehabilitation. We observe that the mean asymmetry in patients was abnormally enhanced across different frequency bands and levels of brain interactions, with these abnormal patterns strongly associated with the side of the stroke lesion. Compared to healthy controls, patients displayed significant alterations in asymmetry fluctuations, disrupting and reconfiguring the balance of inter-hemispheric integration and segregation. Additionally, analyses reveal that specific abnormal asymmetry metrics in patients tend to move towards those observed in healthy controls after short-term brain-computer interface rehabilitation. Furthermore, preliminary evidence suggests that baseline clinical and asymmetry features can predict poststroke improvements in the Fugl-Meyer assessment of the lower extremity (mean absolute error of about 2). Overall, these findings advance our understanding of hemispheric asymmetry. Our framework offers new insights into the mechanisms underlying brain alterations and recovery after a brain lesion, may help identify prognostic biomarkers, and can be easily extended to different functional modalities.

Details DOI

NeurIPS Conference 2023 Conference Paper

Reusing Pretrained Models by Multi-linear Operators for Efficient Training

Yu Pan
Ye Yuan
Yichun Yin
Zenglin Xu
Lifeng Shang
Xin Jiang
Qun Liu

Training large models from scratch usually costs a substantial amount of resources. Towards this problem, recent studies such as bert2BERT and LiGO have reused small pretrained models to initialize a large model (termed the ``target model''), leading to a considerable acceleration in training. Despite the successes of these previous studies, they grew pretrained models by mapping partial weights only, ignoring potential correlations across the entire model. As we show in this paper, there are inter- and intra-interactions among the weights of both the pretrained and the target models. As a result, the partial mapping may not capture the complete information and lead to inadequate growth. In this paper, we propose a method that linearly correlates each weight of the target model to all the weights of the pretrained model to further enhance acceleration ability. We utilize multi-linear operators to reduce computational and spacial complexity, enabling acceptable resource requirements. Experiments demonstrate that our method can save 76\% computational costs on DeiT-base transferred from DeiT-small, which outperforms bert2BERT by +12\% and LiGO by +21\%, respectively.

PDF Details

JBHI Journal 2022 Journal Article

A Transferable Deep Learning Prognosis Model for Predicting Stroke Patients' Recovery in Different Rehabilitation Trainings

Ping-Ju Lin
Xiaoxue Zhai
Wei Li
Tianyi Li
Dandan Cheng
Chong Li
Yu Pan
Linhong Ji

Since the underlying mechanisms of neurorehabilitation are not fully understood, the prognosis of stroke recovery faces significant difficulties. Recovery outcomes can vary when undergoing different treatments; however, few models have been developed to predict patient outcomes toward multiple treatments. In this study, we aimed to investigate the potential of predicting a treatment's outcome using a deep learning prognosis model developed for another treatment. A total of 15 stroke survivors were recruited in this study, and their clinical and physiological data were measured before and after the treatment (clinical measurement, biomechanical measurement, and electroencephalography (EEG) measurement). Multiple biomarkers and clinical scale scores of patients who had completed manual stretching rehabilitation training were analyzed. Data were used to train deep learning prognosis models, yielding an 87. 50% prognosis accuracy. Pre-trained prognosis models were then applied to patients who completed robotic-assisted stretching training, yielding a prognosis accuracy of 91. 84%. Interpretation of the deep learning models revealed several key factors influencing patients' recoveries, including the plantar-flexor active range of movement (r = 0. 930, P = 0. 02), dorsiflexor strength (r = 0. 932, P = 0. 002), plantar-flexor strength (r = 0. 930, P = 0. 002), EEG power spectrum density and EEG functional connectivities in the occipital, central parietal, and parietal areas. Our results suggest (i) that deep learning can be a promising method for accurate prediction of the recovery potential of stroke patients in clinical scenarios and (ii) that it can be successfully applied to different rehabilitation trainings with explainable factors.

Details DOI

YNIMG Journal 2022 Journal Article

Cerebellum anatomy predicts individual risk-taking behavior and risk tolerance

Peng Quan
Lisheng He
Tianxin Mao
Zhuo Fang
Yao Deng
Yu Pan
Xiaocui Zhang
Ke Zhao

Human risk tolerance is highly idiosyncratic and individuals often show distinctive preferences when faced with similar risky situations. However, the neural underpinnings of individual differences in risk-taking remain unclear. Here we combined structural and perfusion MRI and examined the associations between brain anatomy and individual risk-taking behavior/risk tolerance in a sample of 115 healthy participants during the Balloon Analogue Risk Task, a well-established sequential risky decision paradigm. Both whole brain and region-of-interest analyses showed that the left cerebellum gray matter volume (GMV) has a strong association with individual risk-taking behavior and risk tolerance, outperforming the previously reported associations with the amygdala and right posterior parietal cortex (PPC) GMV. Left cerebellum GMV also accounted for risk tolerance and risk-taking behavior changes with aging. However, regional cerebral blood flow (CBF) provided no additional predictive power. These findings suggest a novel cerebellar anatomical contribution to individual differences in risk tolerance. Further studies are necessary to elucidate the underestimated important role of cerebellum in risk-taking.

Details DOI

AAAI Conference 2022 Conference Paper

Making Adversarial Examples More Transferable and Indistinguishable

Junhua Zou
Yexin Duan
Boyu Li
Wu Zhang
Yu Pan
Zhisong Pan

Fast gradient sign attack series are popular methods that are used to generate adversarial examples. However, most of the approaches based on fast gradient sign attack series cannot balance the indistinguishability and transferability due to the limitations of the basic sign structure. To address this problem, we propose a method, called Adam Iterative Fast Gradient Tanh Method (AI-FGTM), to generate indistinguishable adversarial examples with high transferability. Besides, smaller kernels and dynamic step size are also applied to generate adversarial examples for further increasing the attack success rates. Extensive experiments on an ImageNetcompatible dataset show that our method generates more indistinguishable adversarial examples and achieves higher attack success rates without extra running time and resource. Our best transfer-based attack NI-TI-DI-AITM can fool six classic defense models with an average success rate of 89. 3% and three advanced defense models with an average success rate of 82. 7%, which are higher than the state-of-theart gradient-based attacks. Additionally, our method can also reduce nearly 20% mean perturbation. We expect that our method will serve as a new baseline for generating adversarial examples with better transferability and indistinguishability.

PDF Details

YNIMG Journal 2020 Journal Article

EEG microstates are correlated with brain functional networks during slow-wave sleep

Jing Xu
Yu Pan
Shuqin Zhou
Guangyuan Zou
Jiayi Liu
Zihui Su
Qihong Zou
Jia-Hong Gao

Electroencephalography (EEG) microstates have been extensively studied in wakefulness and have been described as the “atoms of thought”. Previous studies of EEG have found four microstates, i. e. , microstates A, B, C and D, that are consistent among participants across the lifespan during the resting state. Studies using simultaneous EEG and functional magnetic resonance imaging (fMRI) have provided evidence for correlations between EEG microstates and fMRI networks during the resting state. Microstates have also been found during non-rapid eye movement (NREM) sleep. Slow-wave sleep (SWS) is considered the most restorative sleep stage and has been associated with the maintenance of sleep. However, the relationship between EEG microstates and brain functional networks during SWS has not yet been investigated. In this study, simultaneous EEG-fMRI data were collected during SWS to test the correspondence between EEG microstates and fMRI networks. EEG microstate-informed fMRI analysis revealed that three out of the four microstates showed significant correlations with fMRI data: 1) fMRI fluctuations in the insula and posterior temporal gyrus positively correlated with microstate B, 2) fMRI signals in the middle temporal gyrus and fusiform gyrus negatively correlated with microstate C, and 3) fMRI fluctuations in the occipital lobe negatively correlated with microstate D, while fMRI signals in the anterior cingulate and cingulate gyrus positively correlated with this microstate. Functional brain networks were then assessed using group independent component analysis based on the fMRI data. The group-level spatial correlation analysis showed that the fMRI auditory network overlapped the fMRI activation map of microstate B, the executive control network overlapped the fMRI deactivation of microstate C, and the visual and salience networks overlapped the fMRI deactivation and activation maps of microstate D. In addition, the subject-level spatial correlations between the general linear model (GLM) beta map of each microstate and the individual maps of each component yielded by dual regression also showed that EEG microstates were closely associated with brain functional networks measured using fMRI during SWS. Overall, the results showed that EEG microstates were closely related to brain functional networks during SWS, which suggested that EEG microstates provide an important electrophysiological basis underlying brain functional networks.

Details DOI

YNIMG Journal 2020 Journal Article

Test-retest reliability of brain responses to risk-taking during the balloon analogue risk task

Xiong Li
Yu Pan
Zhuo Fang
Hui Lei
Xiaocui Zhang
Hui Shi
Ning Ma
Philip Raine

The Balloon Analogue Risk Task (BART) provides a reliable and ecologically valid model for the assessment of individual risk-taking propensity and is frequently used in neuroimaging and developmental research. Although the test-retest reliability of risk-taking behavior during the BART is well established, the reliability of brain activation patterns in response to risk-taking during the BART remains elusive. In this study, we used functional magnetic resonance imaging (fMRI) and evaluated the test-retest reliability of brain responses in 34 healthy adults during a modified BART by calculating the intraclass correlation coefficients (ICC) and Dice’s similarity coefficients (DSC). Analyses revealed that risk-induced brain activation patterns showed good test-retest reliability (median ICC = 0. 62) and moderate to high spatial consistency, while brain activation patterns associated with win or loss outcomes only had poor to fair reliability (median ICC = 0. 33 for win and 0. 42 for loss). These findings have important implications for future utility of the BART in fMRI to examine brain responses to risk-taking and decision-making.

Details DOI

YNIMG Journal 2019 Journal Article

Agreeableness modulates group member risky decision-making behavior and brain activity

Fang Wang
Xin Wang
Fenghua Wang
Li Gao
Hengyi Rao
Yu Pan

When facing difficult decisions, people typically believe that “two heads are better than one”. However, findings from previous studies are inconsistent regarding the advantages of decision-making in groups as compared to individual decision-making. We hypothesize that personality traits may modulate risk-taking behavior and brain activity changes during group decision-making. In this study, we used event-related potentials (ERP) with a well-validated balloon analogue risk task (BART) paradigm to examine the relationships between personality traits, decision-making behavior, and brain activity patterns when a cohort of male participants make decisions and take risks both in groups and in isolation. We found significantly increased risk-taking behavior and reduced P300 component during group decision-making as compared to individual decision-making only for participants with high Agreeableness, but not for those with low Agreeableness. Moreover, Agreeableness scores correlated with risk-taking behavior and P300 amplitude changes in group decisions. These findings suggest that Agreeableness personality modulates risk-taking behavior and brain activity when people make decisions in groups, which have implications for future group decision research and practice.

Details DOI

AAAI Conference 2019 Conference Paper

Compressing Recurrent Neural Networks with Tensor Ring for Action Recognition

Yu Pan
Jing Xu
Maolin Wang
Jinmian Ye
Fei Wang
Kun Bai
Zenglin Xu

Recurrent Neural Networks (RNNs) and their variants, such as Long-Short Term Memory (LSTM) networks, and Gated Recurrent Unit (GRU) networks, have achieved promising performance in sequential data modeling. The hidden layers in RNNs can be regarded as the memory units, which are helpful in storing information in sequential contexts. However, when dealing with high dimensional input data, such as video and text, the input-to-hidden linear transformation in RNNs brings high memory usage and huge computational cost. This makes the training of RNNs very difficult. To address this challenge, we propose a novel compact LSTM model, named as TR-LSTM, by utilizing the low-rank tensor ring decomposition (TRD) to reformulate the input-to-hidden transformation. Compared with other tensor decomposition methods, TR-LSTM is more stable. In addition, TR-LSTM can complete an end-to-end training and also provide a fundamental building block for RNNs in handling large input data. Experiments on real-world action recognition datasets have demonstrated the promising performance of the proposed TR-LSTM compared with the tensor-train LSTM and other state-of-the-art competitors.

PDF Details