A multi-view latent space learning framework via adaptive graph embedding

Haohao Li, Li Zhong

Keywords: Multi-view Learning, Latent representation Learning, Graph Embedding, Graph Optimization

Abstract: In this paper, a new approach to multi-view subspace learning is proposed and termed as multi-view latent space learning via adaptive graph embedding (MvSLGE), which learns a latent representation from all view features. Unlike most existing multi-view latent space learning methods that only encode the complementary information into the latent representation, MvSLGE adaptively learn an adjacent graph that well characterizes similarity between samples to further regularize the latent representation. To extract the neighborhood information from multi-view features, we propose a novel strategy that constructs one graph for each view, and then the learned graph is approximately designed as a centroid of these graphs of different views with different weights. Therefore, the constructed latent representation not only incorporates the complementary information of features from multiple views but also encodes the similarity between samples. The proposed MvSLGE can be solved by the augmented Lagrangian multiplier with alternating direction minimization (ALM-ADM) algorithm. Plenty of experiments demonstrate the superiority of MvSLGE on a variety of datasets.

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/2024/AuthorGuide.

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: 3591