Article
Computer Science, Artificial Intelligence
Yuren Mao, Zekai Wang, Weiwei Liu, Xuemin Lin, Wenbin Hu
Summary: Multi-task Learning (MTL) simultaneously learns multiple tasks for better performance. However, competing tasks can lead to increased task variance, resulting in under-fitting and over-fitting and reduced generalization performance. To address this issue, task variance regularization is proposed as a natural choice for MTL.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Quantum Science & Technology
Shantanav Chakraborty, Aditya Morolia, Anurudh Peduri
Summary: Linear regression is a widely used technique for fitting linear models in machine learning and statistics. However, real-world linear regression problems often suffer from ill-posedness or overfitting, leading to incorrect or trivial solutions. To address this, we propose quantum algorithms for quantum least squares with general e2-regularization, using block-encoding and quantum singular value transformation (QSVT). Our algorithms significantly improve upon previous results in quantum ridge regression and can be applied to various input models, providing improved and generalized versions of standard quantum least squares algorithms.
Article
Engineering, Civil
Zheng Zhu, Meng Xu, Yining Di, Hai Yang
Summary: This paper proposes a physics regularized multi-output grid Gaussian Process Model (PRMGGP) for fast and accurate fitting of large-scale spatial-temporal processes in transportation systems. The PRMGGP model adopts a grid input structure, uses Kronecker algebra for accelerated computation, and incorporates physics laws using a shadow GP. Experimental results demonstrate the efficiency and accuracy of the proposed model.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Mathematics
Xin Wang, Lingchen Kong, Liqun Wang
Summary: In this paper, a novel approach called the natural adaptive lasso is proposed for variance estimation in high-dimensional linear models. This method can simultaneously select and estimate the regression and variance parameters, and its superiority is theoretically proven.
Article
Computer Science, Information Systems
Xiaoxi He, Xu Wang, Zimu Zhou, Jiahang Wu, Zheng Yang, Lothar Thiele
Summary: Future mobile devices are expected to have the ability to perceive, understand, and react to the world independently using deep neural networks. However, these models need to be compressed to fit in mobile storage and memory. This work proposes Multi-Task Zipping (MTZ), a framework that automatically merges pre-trained deep neural networks to reduce redundancy across multiple models. MTZ achieves this through layer-wise neuron sharing and weight updating schemes, which result in minimal changes to the error function. Evaluations show that MTZ effectively merges networks with minimal increase in test errors and significantly reduces the number of iterations required for retraining. It also improves the latency for switching between different tasks on memory-constrained devices.
IEEE TRANSACTIONS ON MOBILE COMPUTING
(2023)
Article
Engineering, Multidisciplinary
Alexandre Cortiella, Kwang-Chun Park, Alireza Doostan
Summary: This work introduces an iterative sparse-regularized regression method for recovering governing equations of nonlinear dynamical systems with improved accuracy and robustness in the presence of state measurement noise. By utilizing a reweighted l(1) regularization approach, the method demonstrates viability for a wide range of potential applications through empirical examples of well-known nonlinear dynamical systems.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2021)
Article
Computer Science, Information Systems
Yanshan Xiao, Liangwang Zhang, Bo Liu, Ruichu Cai, Zhifeng Hao
Summary: In this paper, we propose a novel multi-task semi-supervised ordinal regression (MTSSOR) method to improve the performance of ordinal regression classifiers. Experimental results show that MTSSOR achieves significant improvements compared to other methods in multiple evaluation metrics.
INFORMATION SCIENCES
(2023)
Article
Environmental Sciences
Haoming Zhuang, Xiaoping Liu, Yuchao Yan, Jinpei Ou, Jialyu He, Changjiang Wu
Summary: Fine knowledge of spatiotemporal population distribution is crucial for various fields, but continuous historical multi-temporal gridded population data is lacking. This study developed a framework using deep learning and Landsat-5 images to map multi-temporal population data in China, revealing how population distribution evolved from 1985 to 2010. This framework demonstrates the feasibility of mapping multi-temporal gridded population distribution at a large scale in a timely and cost-effective manner.
Article
Computer Science, Artificial Intelligence
Mohammad Nabati, Seyed Ali Ghorashi, Reza Shahbazian
Summary: This paper introduces a low-cost multi-target Gaussian process regression algorithm, called joint GPR, which employs a shared covariance matrix and solves a sub-optimal cost function for hyperparameter optimization during the training phase. Experimental results show that the proposed method outperforms existing approaches on multiple benchmark datasets.
Article
Computer Science, Artificial Intelligence
Rui Zhang, Hongyuan Zhang, Xuelong Li
Summary: This paper proposes a novel robust multi-task learning model FMC-MTL, which incorporates flexible manifold constraint and robust loss function to handle polluted data. Experimental results demonstrate the remarkable superiority of this model in dealing with outliers.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2021)
Article
Computer Science, Interdisciplinary Applications
Yanjiao Ban, Huan Lao, Bin Li, Wenjun Su, Xuejun Zhang
Summary: In this study, a hypergraph p-Laplacian regularized multi-task feature selection (HpMTFS) method is proposed for the classification of Alzheimer's disease (AD). This method considers the correlation between different modal data and the higher-order relationships between similar data, improving the robustness of the model. The multimodal features are extracted jointly using a group-sparsity regularizer, and a multi-kernel support vector machine is used for fusion and classification.
JOURNAL OF BIOMEDICAL INFORMATICS
(2023)
Article
Mathematics, Applied
Feng Jiang, Lin Du, Fan Yang, Zi-Chen Deng
Summary: This work introduces a regularized least absolute deviation-based sparse identification of dynamics (RLAD-SID) method to handle outlier problems in the classical metric-based loss function and the sparsity constraint framework. The method employs absolute deviation loss as a replacement for Euclidean loss and proposes a computationally efficient optimization algorithm based on the alternating direction method of multipliers. Numerical experiments are conducted to evaluate the effectiveness of RLAD-SID on various nonlinear dynamical systems, and detailed numerical comparisons are provided with other existing metric-based sparse regression methods.
Article
Computer Science, Artificial Intelligence
Bin Li, Zhenqiu Shu, Yingbo Liu, Cunli Mao, Shengxiang Gao, Zhengtao Yu
Summary: Nonnegative matrix factorization (NMF) methods have achieved remarkable performances in multi-view clustering. We propose a novel multi-view learning approach called label-embedded regularized NMF with dual-graph constraints (LeNMF-DC) to obtain a low-dimensional common representation by utilizing prior knowledge hidden in data and constructing three graph regularization terms. Experimental results show that our LeNMF-DC approach outperforms several state-of-the-art approaches in multi-view clustering.
Article
Computer Science, Artificial Intelligence
Lin Feng, Wenzhe Liu, Xiangzhu Meng, Yong Zhang
Summary: This paper introduces an innovative multi-view clustering method, SMCTN, which utilizes triplex regularized non-negative matrix factorization to effectively extract multi-view information while maintaining low-dimensional geometry structure. Extensive experimental results on textual and image datasets demonstrate the superior performance of the proposed method.
Article
Computer Science, Interdisciplinary Applications
Lucas Kook, Torsten Hothorn
Summary: The tramnet package combines flexible transformation models from tram with constrained convex optimization in CVXR to implement regularized linear transformation models. It unifies various regularized regression models under one framework, with features including regularization strategies and hyperparameter optimization. Multiple S3 methods are deployed for visualization, handling, and simulation purposes.