Article
Computer Science, Information Systems
Maria Martinez-Garcia, Pablo Olmos
Summary: The development of high-throughput technologies has led to an increase in the dimensionality of genomics datasets, which poses a challenge for machine learning methods. In this article, the authors propose a deep latent space model for classification and dimensionality reduction, specifically addressing the issues of missing data and limited observations. The proposed model, called Deep Bayesian Logistic Regression (DBLR), produces informative low-dimensional representations, outperforms baseline methods in classification, and can handle missing entries effectively.
IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS
(2023)
Article
Automation & Control Systems
Qingchao Jiang, Jiashi Jiang, Weimin Zhong, Xuefeng Yan
Summary: This study proposes a new Gaussian-process-based probabilistic latent variable (GPPLV) modeling framework for distributed monitoring of multiunit nonlinear processes. The proposed method performs better in showing the nature of different faults and has a higher fault detection rate for large-scale multiunit processes compared to some common distributed process monitoring models.
IEEE TRANSACTIONS ON SYSTEMS MAN CYBERNETICS-SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Feng Xie, Yan Zeng, Zhengming Chen, Yangbo He, Zhi Geng, Kun Zhang
Summary: This paper addresses the challenging problem of causal discovery when the variables of interest cannot be directly measured. By utilizing measurement models, latent variables and their causal relations can be recovered from measured data. The paper provides precise identifiability conditions for the linear pure measurement model and demonstrates the recoverable information of the causal structure from observed data. The effectiveness of the proposed approach is validated through experiments on synthetic and real-world data.
Article
Computer Science, Artificial Intelligence
David Chushig-Muzo, Cristina Soguero-Ruiz, Pablo de Miguel-Bohoyo, Inmaculada Mora-Jimenez
Summary: Electronic health records combined with deep learning methods have provided important outcomes for decision-making, with autoencoders being extensively used in health care. However, the lack of interpretability due to the nonlinear transformation of AE-based models can be addressed by combining probabilistic models and hierarchical clustering to gain insights from AE latent representations.
ARTIFICIAL INTELLIGENCE IN MEDICINE
(2021)
Article
Engineering, Multidisciplinary
Stefania Fresca, Andrea Manzoni
Summary: DL-ROMs are proposed to overcome limitations of conventional ROMs, but require expensive training. The proposed method combines POD and multi-fidelity pretraining to avoid the costly training stage of DL-ROMs.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2022)
Article
Automation & Control Systems
Chunxiao Li, Cynthia Rudin, Tyler H. McCormick
Summary: Instrumental variables are widely used in social and health sciences for causal inference. This paper presents a framework that utilizes machine learning to validate assumptions in the IV model and provides empirical evidence. Prediction validity is the key idea, and one-stage and two-stage approaches for IV are developed.
JOURNAL OF MACHINE LEARNING RESEARCH
(2022)
Article
Computer Science, Artificial Intelligence
Xiang Wang, Junxing Zhu, Zichen Xu, Kaijun Ren, Xinwang Liu, Fengyun Wang
Summary: In this paper, a local nonlinear dimensionality reduction method named Vec2vec is proposed, which utilizes a neural network with one hidden layer to reduce computational complexity. Experimental results demonstrate that Vec2vec outperforms other dimensionality reduction methods in data classification and clustering tasks, and it also requires less computational time in high-dimensional data. Additionally, a lightweight method called Approximate Vec2vec (AVec2vec) is introduced, which achieves competitive performance with UMAP and other local dimensionality reduction methods.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Artificial Intelligence
Guoli Song, Shuhui Wang, Qingming Huang, Qi Tian
Summary: This paper introduces a novel multimodal learning scheme called "Harmonization," which jointly learns latent representations and kernel hyperparameters to address modality heterogeneity. The proposed method outperforms traditional individual learning schemes and shows superior performance in cross-modal retrieval tasks.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2021)
Article
Materials Science, Multidisciplinary
Wenlong Chen, Kai Chen
Summary: The development of material informatics has resulted in a deeper intersection between material science and machine learning. This study proposes a virtual sample generation algorithm using Gaussian process and nonlinear fitting for rubber material. The algorithm is versatile, nonlinear, probabilistic, and interpretable, and it addresses the challenge of limited data volume in materials science.
COMPUTATIONAL MATERIALS SCIENCE
(2023)
Article
Computer Science, Artificial Intelligence
Jie Qiao, Ruichu Cai, Kun Zhang, Zhenjie Zhang, Zhifeng Hao
Summary: The article discusses the identification of causal direction between a causal-effect pair from observed data, proposing a confounding cascade nonlinear additive noise model to address the issue of omitted latent causal variables. Simulation results demonstrate the method's ability to identify indirect causal relations across various settings, while experimental results suggest that the proposed model and method greatly extend the applicability of causal discovery based on functional causal models in nonlinear cases.
ACM TRANSACTIONS ON INTELLIGENT SYSTEMS AND TECHNOLOGY
(2021)
Review
Physics, Multidisciplinary
Andres R. Masegosa, Rafael Cabanas, Helge Langseth, Thomas D. Nielsen, Antonio Salmeron
Summary: Recent advances in statistical inference have expanded the toolbox of probabilistic modeling, allowing for approximate inference on a broad class of models and application to massive data sets. The possibility to incorporate deep neural networks within probabilistic models enables capturing complex non-linear relationships between random variables, greatly expanding the scope of applications.
Article
Engineering, Multidisciplinary
Tiffany Fan, Nathaniel Trask, Marta D'Elia, Eric Darve
Summary: We explore the probabilistic partition of unity network (PPOU-Net) model for high-dimensional regression problems and propose a framework for adaptive dimensionality reduction. The target function is approximated by a mixture of experts model on a low-dimensional manifold, using a training strategy that combines gradient descent and the expectation maximization (EM) algorithm. The PPOU-Nets consistently outperform comparable fully-connected neural networks in various numerical experiments, and the proposed model is also applied in the field of quantum computing.
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING
(2023)
Article
Automation & Control Systems
Changyuan Yang, Sai Ma, Qinkai Han
Summary: This paper proposes a robust discriminant latent variable manifold learning (RDLVML) algorithm for fault diagnosis of rotating machinery. By selecting features of high-dimensional fault data and extracting low-dimensional fault features with better discrimination, the accuracy of fault diagnosis is improved. A novel weighted neighborhood graph is proposed by constructing the q-Rényi and Prime kernel function to suppress the interference of outliers and noise, making the RDLVML algorithm more robust. Furthermore, a fault diagnosis method for rotating machinery based on RDLVML is presented, which achieves more accurate results compared to classical feature selection algorithms through experimental verifications.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Li Wang, Raymond Chan, Tieyong Zeng
Summary: The proposed SSL framework utilizes unlabeled high-dimensional data and a small amount of labeled data to learn a sparse weighted graph, handle data noise, and infer unlabeled data embeddings using a unified model of density estimation and different distance measurements. This approach improves class separation and sparse graph structure learning by estimating density and calculating pairwise distances based on various distance measurements.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2021)
Article
Automation & Control Systems
Angelo Giuseppe Spinosa, Arturo Buscarino, Luigi Fortuna, Matteo Iafrati, Giuseppe Mazzitelli
Summary: The problem of identifying and modeling a complex system requires various techniques and strategies, with computational efforts that can vary significantly. Analyzing the overall complexity of a system is not straightforward due to numerous factors, including the arrangement of constituent items and their interactions. Intuitively, larger sets of sub-parts lead to increased degrees of freedom.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2021)