A novel Visible Multilayer Concept Factorization for Image Data Representation and Clustering
Keywords: Image Data Representation, Concept Factorization, Two-dimensional Feature Extraction, Data Clustering
Abstract: Traditional Concept Factorization (CF) methods learn feature of one data point from high-dimensional data space in the form of vector, which leads to the loss of pixel-level neighborhood information in Two-Dimensional (2D) images. In light of this, we present a novel Visible Multilayer Concept Factorization for image-data representation, termed VMCF. Specifically, to uncover deep latent features from complex data, VMCF adopts a multilayer framework, equipped with a ‘Decomposition, Dimensionality reduction and Data reconstruction’ network ($D^3$-net) in each layer. To obtain locality-preserving features, $D^3$-net firstly performs adaptive graph regularized concept learning on the input data of each layer. Then, $D^3$-net performs 2D feature extraction over the obtained basis images in order to reduce the loss of pixel-level neighborhood information during dimension plunging. The reconstructed data formed by the improved basis images and coefficient matrix is used as input for the next layer. In this way, the dimensions of the original data can gradually decrease at each layer, avoiding information loss caused by sudden dimensionality reduction. Meanwhile, 2D-reduced basis images can mediately improve the quality of new data representations. Extensive numerical experiments on several public image databases have shown that VMCF outperforms other state-of-the-art algorithms.
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 6480
Loading