Article
Computer Science, Artificial Intelligence
Zhibin Gu, Hongzhe Liu, Songhe Feng
Summary: This paper proposes a diversity-induced consensus and structured graph learning model for multi-view clustering, which tackles the issues of consistency and diversity in multiple views and shows good performance in experiments.
APPLIED INTELLIGENCE
(2023)
Review
Computer Science, Artificial Intelligence
Tinghua Wang, Xiaolu Dai, Yuze Liu
Summary: The article provides an in-depth survey of learning methods using the Hilbert-Schmidt independence criterion (HSIC) for various learning problems, such as feature selection, dimensionality reduction, clustering, and kernel learning and optimization. It systematically reviews typical learning models based on the HSIC, ranging from supervised learning to unsupervised learning, traditional machine learning to transfer learning and deep learning. The relationships between learning methods using the HSIC and other relevant learning algorithms are also discussed, aiming to provide practitioners valuable guidelines for their specific domains by elucidating the similarities and differences of these learning models.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Automation & Control Systems
Tianhong Sheng, Bharath K. Sriperumbudur
Summary: This work explores the connection between distance measures and kernel functions in measuring conditional dependence, and finds that in certain cases, distance-based measures and kernel-based measures are equivalent.
JOURNAL OF MACHINE LEARNING RESEARCH
(2023)
Article
Computer Science, Artificial Intelligence
Zhu Li, Adrian Perez-Suay, Gustau Camps-Valls, Dino Sejdinovic
Summary: The current use of machine learning in industrial, societal, and economical activities has raised concerns about the fairness, equity, and ethics of automated decisions. This study presents a regularization approach that balances predictive accuracy with fairness in terms of statistical parity, aiming to address biases in machine learning models.
PATTERN RECOGNITION
(2022)
Review
Engineering, Industrial
Amandine Marrel, Vincent Chabridon
Summary: In this study, target sensitivity analysis (TSA) and conditional sensitivity analysis (CSA) were introduced to better understand the impact of inputs on the output. New operational tools based on Sobol index and HSIC were proposed, along with a method to transform the output using weighted functions to reduce information loss.
RELIABILITY ENGINEERING & SYSTEM SAFETY
(2021)
Article
Computer Science, Artificial Intelligence
Zhi-Chao Sha, Zhang-Meng Liu, Chen Ma, Jun Chen
Summary: The paper introduces a method for feature selection using full-dimensional CMI, successfully achieving relevance maximization and redundance minimization, as demonstrated by experimental results.
APPLIED INTELLIGENCE
(2021)
Article
Computer Science, Artificial Intelligence
Bo Xiong, Hongmei Chen, Tianrui Li, Xiaoling Yang
Summary: Multi-view graph clustering has attracted extensive research attention due to its ability to capture consistent and complementary information between views. However, multi-view data are mostly high-dimensional and may contain redundant and irrelevant features. In addition, the original data are often contaminated by noise and outliers, affecting the reliability of the learned affinity matrix. This study proposes a robust multi-view clustering model that combines low-dimensional and low-rank latent space learning, self-representation learning, and multi-view discrepancy induction fusion. Experimental results on benchmark datasets show that the proposed model outperforms state-of-the-art comparison models in terms of robustness and clustering performance.
APPLIED INTELLIGENCE
(2023)
Article
Engineering, Multidisciplinary
Mohamed Reda El Amri, Amandine Marrel
Summary: This paper proposes new goal-oriented algorithms to optimize the global sensitivity analysis based on the permutation HSIC independence tests. These algorithms include screening-oriented, ranking-oriented, and ranking-screening-oriented types, which have been tested and compared for efficiency and time saving on both analytical examples and thermalhydraulic use cases.
QUALITY AND RELIABILITY ENGINEERING INTERNATIONAL
(2022)
Article
Automation & Control Systems
Hanjia Gao, Xiaofeng Shao
Summary: This paper investigates the behavior of sample MMD in a high-dimensional environment and proposes a new studentized test statistic. Central limit theorems and convergence rates are derived for studentized sample MMD, suggesting that the accuracy of normal approximation improves with dimensionality. The paper also provides a general theory on power analysis and shows the effectiveness of the proposed test in moderately high dimensional regime.
JOURNAL OF MACHINE LEARNING RESEARCH
(2023)
Article
Computer Science, Information Systems
Huibing Wang, Guangqi Jiang, Jinjia Peng, Ruoxi Deng, Xianping Fu
Summary: In this paper, a novel method named multi-view clustering via graph collaboration (MCGC) is proposed to uncover the correlations between multi-view data by collaboratively learning from multiple views. The method utilizes a consensus graph to reveal the essential structure of the data and directly obtains optimal clustering results without postprocessing steps.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Computer Science, Artificial Intelligence
Guohe Li, Yong Li, Yifeng Zheng, Ying Li, Yunfeng Hong, Xiaoming Zhou
Summary: The paper proposes a multi-label feature selection approach called MLFSPO with Pareto optimality, which maps multi-label features to high-dimensional space and evaluates the correlation between features and labels using HSIC. Extensive experimental results on publicly available data sets demonstrate the effectiveness of the proposed algorithm in multi-label tasks.
APPLIED INTELLIGENCE
(2021)
Article
Statistics & Probability
Guochang Wang, Wai Keung Li, Ke Zhu
Summary: This study introduces novel one-sided omnibus tests for independence between two multivariate stationary time series, utilizing the Hilbert-Schmidt independence criterion (HSIC) to analyze the independence between the innovations of the time series. The study establishes the limiting null distributions of the tests under regular conditions and demonstrates the consistency of the HSIC-based tests. The use of a residual bootstrap method for obtaining critical values and the examination of general dependence in contrast to existing linear cross-correlation tests are highlighted as the key contributions of this research.
Article
Computer Science, Information Systems
Zhenxin Wang, Degang Chen, Xiaoya Che
Summary: This paper proposes an effective multi-kernel learning algorithm for multi-label classification by combining local Rademacher complexity and Hilbert-Schmidt independence criterion. The algorithm compresses both the feature space and hypothesis space simultaneously. The experimental results demonstrate the effectiveness of the proposed algorithms in compressing the hypothesis space and feature space.
INFORMATION SCIENCES
(2023)
Article
Biotechnology & Applied Microbiology
Hao Wang, Yijie Ding, Jijun Tang, Quan Zou, Fei Guo
Summary: This study extracts multi-label classification datasets, constructs subcellular localization datasets on different RNA categories, and achieves the identification of multi-label RNA subcellular localizations through multiple kernel learning and support vector machine model, obtaining excellent results.
Article
Computer Science, Artificial Intelligence
Weichao Gan, Zhengming Ma, Shuyu Liu
Summary: This paper introduces a new dimensionality reduction algorithm, PDMHSIC, for tensor data which combines HSIC and tensor algebra. By maximizing HSIC and minimizing the distance between high-dimensional tensor data and their projection in the subspace, this algorithm has outperformed 7 other well-known algorithms on 8 commonly-used datasets in experimental results.
JOURNAL OF INTELLIGENT & FUZZY SYSTEMS
(2021)
Article
Computer Science, Information Systems
Xu Chen, Ya Zhang, Ivor W. Tsang, Yuangang Pan, Jingchao Su
Summary: This article discusses cross-domain recommendation in scenarios where different domains have the same set of users but no overlapping items. Most existing methods focus on shared-user representation, but fail to capture domain-specific features. In this article, an equivalent transformation learner (ETL) is proposed to preserve both domain-specific and overlapped features by modeling the joint distribution of user behaviors across domains.
ACM TRANSACTIONS ON INFORMATION SYSTEMS
(2023)
Article
Engineering, Industrial
Mingxing Li, Daqiang Guo, Ming Li, Ting Qu, George Q. Huang
Summary: The widespread adoption of Industry 4.0 technologies is revolutionising manufacturing operations. This paper introduces a novel concept of operations twins (OT) for achieving synchronisation between production and intralogistics (PiL) through the use of Industry 4.0 technologies and innovative operations management strategies.
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH
(2023)
Article
Computer Science, Artificial Intelligence
Xiaowei Zhou, Ivor W. Tsang, Jie Yin
Summary: Deep Neural Networks have achieved great success in classification tasks, but they are vulnerable to adversarial attacks. Adversarial training is an effective strategy to improve the robustness of DNN models, but existing methods fail to generalize well to standard test data. To achieve a better trade-off between standard accuracy and adversarial robustness, a novel adversarial training framework called LADDER is proposed, which generates high-quality adversarial examples through perturbations on latent features.
Article
Computer Science, Artificial Intelligence
Peiyao Zhao, Yuangang Pan, Xin Li, Xu Chen, Ivor W. Tsang, Lejian Liao
Summary: Inspired by the success of contrastive learning, graph augmentation strategies have been used to learn node representations. Existing methods add perturbations to construct contrastive samples. However, they ignore the prior information assumption, leading to decreased similarity and increased discrimination among nodes. In this article, a general ranking framework is proposed to incorporate these prior information into contrastive learning. Experimental results on benchmark datasets show the effectiveness of the proposed algorithm compared to supervised and unsupervised models.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Jiangchao Yao, Bo Han, Zhihan Zhou, Ya Zhang, Ivor W. Tsang
Summary: Learning with noisy labels is important in the Big Data era to save costs. Previous noise-transition-based methods achieved good performance but relied on impractical anchor sets. Our approach introduces a Bayesian framework for parameterizing the noise transition and solves the problem of ill-posed stochastic learning in back-propagation.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Bing Li, Wei Cui, Le Zhang, Ce Zhu, Wei Wang, Ivor W. Tsang, Joey Tianyi Zhou
Summary: Time series analysis is crucial in various fields such as economics, finance, and surveillance. However, traditional Transformer models have limitations in representing nuanced patterns in time series data. To overcome these challenges, we propose a novel Transformer architecture called DifFormer, which incorporates a multi-resolutional differencing mechanism. DifFormer outperforms existing models in classification, regression, and forecasting tasks, while also exhibiting efficiency and lower time consumption.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Defu Liu, Wen Li, Lixin Duan, Ivor W. Tsang, Guowu Yang
Summary: Deep models have achieved impressive performance in various visual recognition tasks, but their generalization ability is compromised by noisy labels. This paper presents a dynamic label learning algorithm that allows the use of different loss functions for classification in the presence of label noise, ensuring that the search for the optimal classifier of noise-free samples is not hindered by label noise.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Yongshun Gong, Zhibin Li, Wei Liu, Xiankai Lu, Xinwang Liu, Ivor W. W. Tsang, Yilong Yin
Summary: Many real-world problems involve data with missing values, which can hinder learning achievements. Existing methods use a universal model for all incomplete data, resulting in suboptimal models for each missingness pattern. This paper proposes a general model that can adjust to different missingness patterns, minimizing competition between data. The model is based on observable features and does not rely on data imputation, and a low-rank constraint is introduced to improve generalization ability.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Hui Xu, Changyu Li, Yan Zhang, Lixin Duan, Ivor W. Tsang, Jie Shao
Summary: Cold-start is crucial for recommendations, especially when there is limited user-item interaction data. Meta-learning-based approaches have shown success in addressing this issue by leveraging their strong generalization capabilities to quickly adapt to new tasks in cold-start settings. However, these methods are prone to meta-overfitting when trained with single and sparse ratings because a single rating cannot capture a user's diverse interests under different circumstances. To overcome this, a meta-augmentation technique is proposed to convert non-mutually-exclusive (Non-ME) tasks into mutually-exclusive (ME) tasks without changing inputs, thus relieving the issue of meta-overfitting. Inspired by this technique, this paper proposes a cross-domain meta-augmentation technique for content-aware recommendation systems (MetaCAR) to construct ME tasks in the recommendation scenario. The proposed method consists of two stages: meta-augmentation and meta-learning. In the meta-augmentation stage, domain adaptation is performed using a dual conditional variational autoencoder (CVAE) with a multi-view information bottleneck constraint, and the learned CVAE is applied to generate ratings for users in the target domain. In the meta-learning stage, both the true and generated ratings are used to construct ME tasks, enabling meta-learning recommendations to avoid meta-overfitting. Experiments conducted on real-world datasets demonstrate the significant superiority of MetaCAR over competing baselines, including cross-domain, content-aware, and meta-learning-based recommendations, in dealing with the cold-start user issue.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Jinliang Deng, Xiusi Chen, Renhe Jiang, Xuan Song, Ivor W. Tsang
Summary: Multi-variate time series (MTS) data is a common type of data abstraction in the real world, generated from a hybrid dynamical system. MTS data can be categorized into spatial and temporal attributes, and can be analyzed from the spatial view or temporal view. A novel multi-view multi-task (MVMT) learning framework is proposed to extract hidden MVMT information from MTS data while predicting. The framework improves effectiveness and efficiency of canonical architectures according to extensive experiments on three datasets.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Shudong Huang, Ivor W. W. Tsang, Zenglin Xu, Jiancheng Lv
Summary: Multi-view clustering aims to reveal correlations between different input modalities in an unsupervised way. This paper proposes a novel model that learns a robust structured similarity graph and performs multi-view clustering simultaneously. The similarity graph is adaptively learned based on a latent representation that is invulnerable to noise and outliers. Experimental results on benchmark datasets demonstrate the effectiveness of the proposed model.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Jing Li, Yuangang Pan, Ivor W. Tsang
Summary: This article proposes a dual mechanism called adaptive sharpening (ADS) to minimize prediction uncertainty in semi-supervised learning. ADS applies a soft-threshold to mask out uncertain and negligible predictions, and sharpens the informed ones to distill certain predictions.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Xiaofeng Cao, Ivor W. Tsang
Summary: Machine teaching is a reverse problem of machine learning, aiming to guide the student towards its target hypothesis using known learning parameters. Previous studies focused on balancing teaching risk and cost to find the best teaching examples. However, when the student doesn't disclose any cue of the learning parameters, the optimization solver becomes ineffective. This article presents a distribution matching-based machine teaching strategy that iteratively shrinks teaching cost to eliminate boundary perturbations, providing an effective solution.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Yuhao Liu, Qing Guo, Lan Fu, Zhanghan Ke, Ke Xu, Wei Feng, Ivor W. Tsang, Rynson W. H. Lau
Summary: In this paper, a novel structure-informed shadow removal network (StructNet) is proposed to address the problem of shadow remnants in existing deep learning-based methods. StructNet reconstructs the structure information of the input image without shadows and uses it to guide the image-level shadow removal. Two main modules, MSFE and MFRA, are developed to extract image structural features and regularize feature consistency. Additionally, an extension called MStructNet is proposed to exploit multi-level structure information and improve shadow removal performance.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Xiaofeng Cao, Weiyang Liu, Ivor W. Tsang
Summary: This paper addresses the problem of data-efficient learning from scratch in scenarios where data or labels are expensive to collect. It proposes the MHEAL algorithm based on active learning on homeomorphic tubes of spherical manifolds, and provides comprehensive theoretical guarantees. Empirical results demonstrate the effectiveness of MHEAL in various applications for data-efficient learning.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
