Article
Engineering, Electrical & Electronic
Arpan Mukherjee, Ali Tajer, Pin-Yu Chen, Payel Das
Summary: This paper focuses on a sequential estimation approach for estimating shared and private parameters of K processes. The proposed active sampling algorithm makes data-driven sampling decisions and provides estimators for the parameters, achieving reliable estimates with the fewest number of samples.
IEEE TRANSACTIONS ON SIGNAL PROCESSING
(2022)
Article
Computer Science, Artificial Intelligence
Deborah Pereg
Summary: The statistical supervised learning framework assumes a joint probability distribution that can be accurately represented by the training dataset. This work investigates the relationship between sample complexity, empirical risk, and generalization error based on the asymptotic equipartition property. The study provides theoretical guarantees for reliable learning in different settings regarding generalization error and sample size.
Article
Automation & Control Systems
Yagiz Savas, Vijay Gupta, Ufuk Topcu
Summary: This article discusses the problem of synthesizing incentives to induce desired agent behavior when the agent's intrinsic motivation is unknown. The agent's behavior is modeled as a Markov decision process, and linear programming is used to solve the problem.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL
(2022)
Article
Engineering, Electrical & Electronic
Boshuang Huang, Sudeep Salgia, Qing Zhao
Summary: This article studies online active learning for classifying streaming instances within the framework of statistical learning theory. By developing a disagreement-based online learning algorithm and establishing the tradeoff between label complexity and regret, an optimized algorithm is proposed.
IEEE TRANSACTIONS ON SIGNAL PROCESSING
(2022)
Article
Computer Science, Artificial Intelligence
Chi Zhang, Benyi Hu, Yuhang Liuzhang, Le Wang, Li Liu, Yuehu Liu
Summary: Long-tailed visual recognition poses challenges to traditional machine learning and deep networks. Existing methods lack theory and fail to solve the paradoxical effects of long tail. This paper proposes a principled solution and a sampling strategy called Switching, which achieves more efficient performance in long-tailed learning.
Article
Computer Science, Information Systems
Min Cui, Yang Liu, Yanbo Wang, Pan Wang
Summary: Acoustic signal classification is crucial for acoustic source identification, but limited training data often leads to low sample complexity. This study proposes a data fusion model, MFF-ResNet, that combines manual design features and deep representation of log-Mel spectrogram features, along with prior human knowledge as implicit regularization, resulting in a low sample complexity model for accurate acoustic signal classification.
Article
Computer Science, Artificial Intelligence
Luciano Henrique Peixoto da Silva, Lucas Henrique Sousa Mello, Alexandre Rodrigues, Flavio Miguel Varejao, Marcos Pellegrini Ribeiro, Thiago Oliveira-Santos
Summary: This paper proposes an intelligent fault diagnosis method for Electrical Submersible Pump using uncertainty-based active learning to assist experts in labeling data and searching for samples of new fault types. By analyzing features from vibration signals, the proposed approach tests classical classification algorithms and deep learning methods, and introduces a new acquisition strategy for active learning to improve the performance of feature extractors.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Statistics & Probability
Joe Marion, Joseph Mathews, Scott C. Schmidler
Summary: We present bounds for the finite-sample error of sequential Monte Carlo samplers on static spaces. Our approach explicitly relates the performance of the algorithm to properties of the chosen sequence of distributions and mixing properties of the associated Markov kernels. This allows us to give the first finite-sample comparison to other Monte Carlo schemes. We obtain bounds for the complexity of sequential Monte Carlo approximations for a variety of target distributions such as finite spaces, product measures, and log-concave distributions including Bayesian logistic regression. The bounds obtained are within a logarithmic factor of similar bounds obtainable for Markov chain Monte Carlo.
ANNALS OF STATISTICS
(2023)
Article
Engineering, Civil
Chengning Zhou, Lingjie Wang, Yuqi Chen
Summary: This paper proposes a novel method, PAK-SEIS, which combines the Parallel Active learning Kriging model and the Sequential Importance Sampling method to efficiently analyze the structural system reliability with multiple failure modes and small failure probabilities. The method includes a new sequential importance sampling method that integrates sequential Monte Carlo simulation and kernel density estimation. The proposed parallel learning strategy allows for the selection of multiple new training samples and reduces the iterations of Kriging models.
Article
Mathematics, Applied
Max Ehre, Iason Papaioannou, Bruno Sudret, Daniel Straub
Summary: This study addresses the challenge of analyzing high-dimensional, computationally expensive engineering models in risk and reliability engineering using a combination of dimensionality reduction and surrogate modeling. The approach is extended with an active learning procedure to improve error control. The performance of this approach is demonstrated with various example problems featuring well-known caveats for reliability methods.
SIAM JOURNAL ON SCIENTIFIC COMPUTING
(2022)
Article
Mathematics, Applied
Max Ehre, Iason Papaioannou, Bruno Sudret, Daniel Straub
Summary: This paper presents a method combining dimensionality reduction and surrogate modeling to address the analysis of high-dimensional, computationally expensive engineering models. Through an active learning procedure, improved error control can be achieved at each importance sampling level.
SIAM JOURNAL ON SCIENTIFIC COMPUTING
(2022)
Article
Computer Science, Information Systems
Sekjin Hwang, Jinwoo Choi, Joonsoo Choi
Summary: The paper introduces the uncertainty-based Selective Clustering for Active Learning (SCAL) method, which selectively clusters data with high uncertainty to reduce redundancy, thus extending the area of the decision boundary represented by the sampled data. SCAL achieves cutting-edge performance for classification tasks on balanced and unbalanced image datasets as well as semantic segmentation tasks.
Article
Computer Science, Information Systems
Fan Li, Bo Wang, Yinghua Shen, Pin Wang, Yongming Li
Summary: This paper proposes an imbalanced ensemble learning algorithm based on weighted projection clustering grouping and consistent fuzzy sample transformation. It utilizes a weighted projection clustering combination framework to obtain high-quality clusters and applies a stage-wise hybrid sampling algorithm for de-overlapping and balancing of subsets. Additionally, a local-global structure consistency mechanism is constructed to improve the quality of samples in subsets. Experimental results demonstrate the superiority of the proposed algorithm in terms of anti-overlapping, Recall, F1-M, G-M, AUC, and diversity.
INFORMATION SCIENCES
(2023)
Article
Computer Science, Information Systems
Kyoungok Kim
Summary: The study introduced new hybrid sampling/ensemble algorithms, NASBoost and NASBagging, based on a modification of SMOTE, which improved classification performance by preventing the generation of noise in the minority class while maintaining diversity among training sets.
Article
Engineering, Multidisciplinary
Li Wang, Xin Ye, Jialin Li, Yu Wen, Wenbin Liao, Houbing Song, Jie Chen, Jianqiang Li
Summary: In this paper, an architecture utilizing generative adversarial networks and dual active learning modules was proposed to address the issues of imbalanced and scarce data in hospital acquired infections detection. The results showed that this approach improved accuracy and F1-score, demonstrating its effectiveness and efficiency.
IEEE TRANSACTIONS ON NETWORK SCIENCE AND ENGINEERING
(2022)