Article
Chemistry, Analytical
Xuejun Zhao, Rafal Stanislawski, Paolo Gardoni, Maciej Sulowicz, Adam Glowacz, Grzegorz Krolczyk, Zhixiong Li
Summary: In this paper, we discuss the source-free unsupervised domain adaptation problem and propose a label consistent contrastive learning (LCCL) method. Extensive experiments on digit recognition and image classification datasets demonstrate the effectiveness of the proposed method.
Article
Engineering, Biomedical
Xuan Zhang, Xu Zhang, Le Wu, Chang Li, Xiang Chen, Xun Chen
Summary: This article introduces a gesture recognition method based on unsupervised domain adaptation, which utilizes a self-guided adaptive sampling strategy to improve the feature representation consistency of myoelectric patterns across users. Experimental results demonstrate the excellent performance of this method in cross-user classification.
IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING
(2022)
Article
Engineering, Electrical & Electronic
Hengyang Wang, Xianghao Zhan, Li Liu, Asif Ullah, Huiyan Li, Han Gao, You Wang, Ruifen Hu, Guang Li
Summary: This study investigates the improvement of generalizability and transferability of taste sensation models developed with sEMG data by innovatively applying two methods: domain regularized component analysis (DRCA) and conformal prediction with shrunken centroids (CPSC). The effectiveness of these methods is explored in an unlabeled data augmentation process on six subjects. The results show that DRCA significantly improves classification accuracy, while CPSC does not guarantee accuracy improvement. The combination of DRCA and CPSC presents a statistically significant improvement in classification accuracy on six subjects.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2022)
Article
Computer Science, Artificial Intelligence
Luxin Zhang, Pascal Germain, Yacine Kessaci, Christophe Biernacki
Summary: We study a realistic domain adaptation setting where an existing black-box machine learning model is accessible. We propose a solution that provides an interpretable target to source transformation by adapting the feature space in a sparse and ordered manner. The selection of features to be adapted is done using a weakly-supervised process and a new pseudo-label estimator based on rank-stability. Experimental results show promising performance on real datasets.
Article
Engineering, Biomedical
Weihai Chen, Mingxing Lyu, Xilun Ding, Jianhua Wang, Jianbin Zhang
Summary: This paper presents an EMG-based gait pattern adaptation method that allows subjects to control a robotic exoskeleton for gait rehabilitation. The results show that the subjects were able to change the gait pattern of the exoskeleton using EMG signals and achieved the adaptation goals within a short period of time. This method enables subjects to actively participate in the rehabilitation training.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2023)
Article
Computer Science, Artificial Intelligence
Weikai Li, Songcan Chen
Summary: This paper proposes a novel unsupervised domain adaptation method named PAS, which gradually obtains reliable pseudo labels to mitigate mode collapse. The method not only performs well in common UDA scenarios, but also outperforms state-of-the-art methods in the more challenging partial domain adaptation situation where the source label set subsumes the target one.
PATTERN RECOGNITION
(2022)
Article
Computer Science, Artificial Intelligence
Qing Tian, Chuang Ma, Meng Cao, Jun Wan, Zhen Lei, Songcan Chen
Summary: Unsupervised domain adaptation (UDA) is a learning paradigm that utilizes labeled source domain knowledge to improve the unlabeled target domain. This article proposes a dictionary learning-based unsupervised multitarget domain adaptation method (DL-UMTDA) that constructs a common dictionary and individual dictionaries to exploit the relationships between the source and target domains, as well as among the target domains.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Multidisciplinary Sciences
Hyunsik Jeon, Seongmin Lee, U. Kang
Summary: This study addresses the problem of unsupervised multi-source domain adaptation without observable source data by proposing a novel architecture called DEM (Data-free Exploitation of Multiple Sources), which adapts target data to source domains and estimates target labels by exploiting pre-trained source classifiers. Extensive experiments have shown that DEM provides state-of-the-art accuracy for data-free UMDA on real-world datasets, achieving up to 27.5% higher accuracy compared to the best baseline.
Article
Chemistry, Analytical
Jose L. Gomez, Gabriel Villalonga, Antonio M. Lopez
Summary: This paper proposes a new co-training procedure for the unsupervised domain adaptation of semantic segmentation models from synthetic to real images. The procedure involves training intermediate deep models with both synthetic and real images and iteratively labeling real-world training images. The collaboration between the models is achieved through a self-training stage and a model collaboration loop. Experimental results demonstrate significant improvements over baselines on standard synthetic and real-world datasets.
Article
Computer Science, Artificial Intelligence
Xin Luo, Wei Chen, Zhengfa Liang, Chen Li, Yusong Tan
Summary: This study proposes a new unsupervised domain adaptation method that achieves fine-grained domain adaptation by differentiating individual samples and introducing a style discrepancy metric. Experimental results validate the effectiveness of the proposed method, which outperforms many existing adversarial-learning-based methods on different tasks.
Article
Computer Science, Artificial Intelligence
Song Tang, Yan Zou, Zihao Song, Jianzhi Lyu, Lijuan Chen, Mao Ye, Shouming Zhong, Jianwei Zhang
Summary: This study proposes a new source data-free unsupervised domain adaptation method, which addresses the limitations of existing methods in defining geometric structures and depicting semantic relationships by conducting semantic consistency learning on a manifold.
Article
Computer Science, Information Systems
Mengmeng Jing, Lichao Meng, Jingjing Li, Lei Zhu, Heng Tao Shen
Summary: We propose a novel adversarial domain adaptation method called AMRC to reduce domain shifts. Our method uses mixup to generate multiple features with different mixup ratios, and learns a continuous and domain-invariant latent space by accurately estimating the mixup ratio and making the estimator uncertain about it. We also apply mixup regularizations to ensure the smoothness and continuity of the latent space, and enhance the discriminability of the target features through self-supervised learning using sharpened pseudo-labels. Experimental results on 3 benchmarks validate the effectiveness of our method.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Computer Science, Information Systems
Indu Joshi, Tushar Prakash, Rohit Kumar, Antitza Dantcheva, Sumantra Dutta Roy, Prem Kumar Kalra
Summary: This research proposes a method to improve the generalization of fingerprint denoising models by aligning the synthetic and real fingerprint domains. Experimental results show that after domain alignment, the error rate of fingerprint recognition decreases significantly.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Chemistry, Multidisciplinary
Pablo Gimeno, Dayana Ribas, Alfonso Ortega, Antonio Miguel, Eduardo Lleida
Summary: This paper evaluates three unsupervised domain adaptation techniques in the context of Speech Activity Detection (SAD) and identifies effective methods to improve the performance of the SAD task.
APPLIED SCIENCES-BASEL
(2022)
Article
Multidisciplinary Sciences
Mohammad Fazle Rabbi, Laura E. Diamond, Chris P. Carty, David G. Lloyd, Giorgio Davico, Claudio Pizzolato
Summary: This study aims to develop a method to estimate activation patterns of lower limb muscles from electromyograms (EMG) measured from a small set of muscles in children with cerebral palsy. The results show that our muscle synergy extrapolation method can estimate the activation patterns of unmeasured muscles in children with cerebral palsy using only three to four experimental EMG.
SCIENTIFIC REPORTS
(2022)
Article
Computer Science, Artificial Intelligence
Xiao Gu, Yao Guo, Fani Deligianni, Benny Lo, Guang-Zhong Yang
Summary: For abnormal gait recognition, the integration of pattern-specific features and subject-specific differences in deep representations can lead to overfitting and hinder generalization to new subjects. Limited availability of abnormal gait data from precise Motion Capture systems is compounded by slow adaptation of new technologies in healthcare. Our proposed cascade of deep architectures aims to address these challenges by encoding cross-modal and cross-subject transfer for improved abnormal gait recognition accuracy.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Yao Guo, Daniel Freer, Fani Deligianni, Guang-Zhong Yang
Summary: Monitoring mental workload of operators in space telerobotic training and teleoperation tasks is crucial. This study investigates the impact of time-pressure and latency on space teleoperation, using eye-tracking technology for mental workload estimation and performance evaluation. Significant eye-tracking features such as fixation duration, saccade frequency and duration, pupil diameter, and pupillary activity index are identified for mental workload estimation and task performance evaluation.
IEEE TRANSACTIONS ON HUMAN-MACHINE SYSTEMS
(2022)
Article
Computer Science, Information Systems
Xiao Gu, Yao Guo, Guang-Zhong Yang, Benny Lo
Summary: Accurate lower-limb pose estimation is crucial for skeleton based pathological gait analysis. Existing methods have limitations, and we propose a novel cross-domain self-supervised learning framework to address these issues and achieve accurate and precise pose estimation.
IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS
(2022)
Article
Robotics
Jianing Qiu, Lipeng Chen, Xiao Gu, Frank P-W Lo, Ya-Yen Tsai, Jiankai Sun, Jiaqi Liu, Benny Lo
Summary: In this study, the problem of forecasting the trajectory of an egocentric camera wearer in crowded spaces is addressed. A novel egocentric human trajectory forecasting dataset is constructed, and a Transformer-based neural network model integrated with a cascaded cross-attention mechanism is designed. The results show that the proposed model outperforms the state-of-the-art methods in egocentric human trajectory forecasting.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2022)
Article
Computer Science, Artificial Intelligence
Hanxiao Zhang, Liang Chen, Xiao Gu, Minghui Zhang, Yulei Qin, Feng Yao, Zhexin Wang, Yun Gu, Guang-Zhong Yang
Summary: This study addresses the generalizability issues in lung nodule classification by constructing a sure-annotation dataset and proposing a collaborative learning framework. By integrating unsure-annotation data knowledge through nodule segmentation and malignancy score regression, the proposed approach achieves improved performance and trustworthy model reasoning.
MEDICAL IMAGE ANALYSIS
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Hanxiao Zhang, Liang Chen, Minghui Zhang, Xiao Gu, Yulei Qin, Weihao Yu, Feng Yao, Zhexin Wang, Yun Gu, Guang-Zhong Yang
Summary: Accurate nodule labeling and interpretable machine learning play crucial roles in lung cancer diagnosis. The proposed collaborative model and regularization strategy combination achieves the best performances in lung cancer prediction and interpretable diagnosis.
INTERPRETABILITY OF MACHINE INTELLIGENCE IN MEDICAL IMAGE COMPUTING, IMIMIC 2022
(2022)
Proceedings Paper
Automation & Control Systems
Yuxuan Liu, Jianxin Yang, Xiao Gu, Yao Guo, Guang-Zhong Yang
Summary: This paper proposes a system based on egocentric vision for 3D canonical pose estimation and human-centric social interaction characterization. By leveraging global context and two head-mounted cameras, the system can accurately estimate poses and handle various social interaction tasks.
2022 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS (IROS)
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Xiao Gu, Yao Guo, Zeju Li, Jianing Qiu, Qi Dou, Yuxuan Liu, Benny Lo, Guang-Zhong Yang
Summary: This study investigates the problems of long-tailed classification and domain shift, proposing three new core functional blocks and adopting a meta-learning framework to improve domain generalization on unseen target domains. Experimental results demonstrate superior performance over existing approaches for long-tailed/domain generalization tasks.
COMPUTER VISION, ECCV 2022, PT XXIII
(2022)
Proceedings Paper
Computer Science, Information Systems
Jiachuan Peng, Peilun Shi, Jianing Qiu, Xinwei Ju, Frank P. -W. Lo, Xiao Gu, Wenyan Jia, Tom Baranowski, Matilda Steiner-Asiedu, Alex K. Anderson, Megan A. McCrory, Edward Sazonov, Mingui Sun, Gary Frost, Benny Lo
Summary: In this study, researchers collected a large number of in-the-wild images using wearable camera technologies in Ghana. They proposed a novel self-supervised learning framework to cluster the images into separate events, which enabled more efficient data analysis. The framework outperformed baseline methods in terms of clustering quality and classification accuracy.
2022 IEEE-EMBS INTERNATIONAL CONFERENCE ON BIOMEDICAL AND HEALTH INFORMATICS (BHI) JOINTLY ORGANISED WITH THE IEEE-EMBS INTERNATIONAL CONFERENCE ON WEARABLE AND IMPLANTABLE BODY SENSOR NETWORKS (BSN'22)
(2022)
Proceedings Paper
Computer Science, Interdisciplinary Applications
Saeed Akbarzadeh, Xiao Gu, Zhipeng Wu, Benny Lo
Summary: The article introduces a wearable device that utilizes echolocation technique to help blind and visually impaired individuals acquire navigation skills. The research results show that trained subjects can 'visualize' the surrounding environment using the echoed signals.
2022 IEEE-EMBS INTERNATIONAL CONFERENCE ON BIOMEDICAL AND HEALTH INFORMATICS (BHI'22) JOINTLY ORGANISED WITH THE IEEE-EMBS INTERNATIONAL CONFERENCE ON WEARABLE AND IMPLANTABLE BODY SENSOR NETWORKS (BSN'22)
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Hanxiao Zhang, Xiao Gu, Minghui Zhang, Weihao Yu, Liang Chen, Zhexin Wang, Feng Yao, Yun Gu, Guang-Zhong Yang
Summary: The LIDC-IDRI database, a popular benchmark for lung cancer prediction, may have label assignment errors due to subjective assessment. This study proposes a strategy of re-labeling LIDC data using similar nodule retrieval, which improves model performance.
MEDICAL IMAGE LEARNING WITH LIMITED AND NOISY DATA (MILLAND 2022)
(2022)
Proceedings Paper
Automation & Control Systems
Jianing Qiu, Frank P-W Lo, Xiao Gu, Yingnan Sun, Shuo Jiang, Benny Lo
Summary: This research constructed a new egocentric dataset using a wearable camera and designed an LSTM-based encoder-decoder framework to predict the future location and movement trajectory of the targeted person. Experimental results have shown that the proposed method is more reliable and accurate in predicting the future person location and trajectory in egocentric videos compared to three baselines.
2021 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS (IROS)
(2021)
Article
Computer Science, Artificial Intelligence
Xiao Gu, Yao Guo, Fani Deligianni, Guang-Zhong Yang
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2020)