{"title":"MDSR-NMF:用于非负矩阵分解的多重解构单重构深度神经网络模型。","authors":"Prasun Dutta, Rajat K De","doi":"10.1080/0954898X.2023.2257773","DOIUrl":null,"url":null,"abstract":"<p><p>Dimension reduction is one of the most sought-after strategies to cope with high-dimensional ever-expanding datasets. To address this, a novel deep-learning architecture has been designed with multiple deconstruction and single reconstruction layers for non-negative matrix factorization aimed at low-rank approximation. This design ensures that the reconstructed input matrix has a unique pair of factor matrices. The two-stage approach, namely, pretraining and stacking, aids in the robustness of the architecture. The sigmoid function has been adjusted in such a way that fulfils the non-negativity criteria and also helps to alleviate the data-loss problem. Xavier initialization technique aids in the solution of the exploding or vanishing gradient problem. The objective function involves regularizer that ensures the best possible approximation of the input matrix. The superior performance of MDSR-NMF, over six well-known dimension reduction methods, has been demonstrated extensively using five datasets for classification and clustering. Computational complexity and convergence analysis have also been presented to establish the model.</p>","PeriodicalId":54735,"journal":{"name":"Network-Computation in Neural Systems","volume":null,"pages":null},"PeriodicalIF":1.1000,"publicationDate":"2023-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"MDSR-NMF: Multiple deconstruction single reconstruction deep neural network model for non-negative matrix factorization.\",\"authors\":\"Prasun Dutta, Rajat K De\",\"doi\":\"10.1080/0954898X.2023.2257773\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p><p>Dimension reduction is one of the most sought-after strategies to cope with high-dimensional ever-expanding datasets. To address this, a novel deep-learning architecture has been designed with multiple deconstruction and single reconstruction layers for non-negative matrix factorization aimed at low-rank approximation. This design ensures that the reconstructed input matrix has a unique pair of factor matrices. The two-stage approach, namely, pretraining and stacking, aids in the robustness of the architecture. The sigmoid function has been adjusted in such a way that fulfils the non-negativity criteria and also helps to alleviate the data-loss problem. Xavier initialization technique aids in the solution of the exploding or vanishing gradient problem. The objective function involves regularizer that ensures the best possible approximation of the input matrix. The superior performance of MDSR-NMF, over six well-known dimension reduction methods, has been demonstrated extensively using five datasets for classification and clustering. Computational complexity and convergence analysis have also been presented to establish the model.</p>\",\"PeriodicalId\":54735,\"journal\":{\"name\":\"Network-Computation in Neural Systems\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":1.1000,\"publicationDate\":\"2023-02-01\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Network-Computation in Neural Systems\",\"FirstCategoryId\":\"94\",\"ListUrlMain\":\"https://doi.org/10.1080/0954898X.2023.2257773\",\"RegionNum\":3,\"RegionCategory\":\"计算机科学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"2023/11/9 0:00:00\",\"PubModel\":\"Epub\",\"JCR\":\"Q4\",\"JCRName\":\"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Network-Computation in Neural Systems","FirstCategoryId":"94","ListUrlMain":"https://doi.org/10.1080/0954898X.2023.2257773","RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"2023/11/9 0:00:00","PubModel":"Epub","JCR":"Q4","JCRName":"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE","Score":null,"Total":0}
MDSR-NMF: Multiple deconstruction single reconstruction deep neural network model for non-negative matrix factorization.
Dimension reduction is one of the most sought-after strategies to cope with high-dimensional ever-expanding datasets. To address this, a novel deep-learning architecture has been designed with multiple deconstruction and single reconstruction layers for non-negative matrix factorization aimed at low-rank approximation. This design ensures that the reconstructed input matrix has a unique pair of factor matrices. The two-stage approach, namely, pretraining and stacking, aids in the robustness of the architecture. The sigmoid function has been adjusted in such a way that fulfils the non-negativity criteria and also helps to alleviate the data-loss problem. Xavier initialization technique aids in the solution of the exploding or vanishing gradient problem. The objective function involves regularizer that ensures the best possible approximation of the input matrix. The superior performance of MDSR-NMF, over six well-known dimension reduction methods, has been demonstrated extensively using five datasets for classification and clustering. Computational complexity and convergence analysis have also been presented to establish the model.
期刊介绍:
Network: Computation in Neural Systems welcomes submissions of research papers that integrate theoretical neuroscience with experimental data, emphasizing the utilization of cutting-edge technologies. We invite authors and researchers to contribute their work in the following areas:
Theoretical Neuroscience: This section encompasses neural network modeling approaches that elucidate brain function.
Neural Networks in Data Analysis and Pattern Recognition: We encourage submissions exploring the use of neural networks for data analysis and pattern recognition, including but not limited to image analysis and speech processing applications.
Neural Networks in Control Systems: This category encompasses the utilization of neural networks in control systems, including robotics, state estimation, fault detection, and diagnosis.
Analysis of Neurophysiological Data: We invite submissions focusing on the analysis of neurophysiology data obtained from experimental studies involving animals.
Analysis of Experimental Data on the Human Brain: This section includes papers analyzing experimental data from studies on the human brain, utilizing imaging techniques such as MRI, fMRI, EEG, and PET.
Neurobiological Foundations of Consciousness: We encourage submissions exploring the neural bases of consciousness in the brain and its simulation in machines.