Semantically consistent multi-view representation learning

In this work, we devote ourselves to the challenging task of Unsupervised Multi-view Representation Learning (UMRL), which requires learning a unified feature representation from multiple views in an unsupervised manner. Existing UMRL methods mainly focus on process within space while ignoring valuable semantic information hidden different views. To address issue, propose novel approach called Semantically Consistent (SCMRL), aims excavate underlying multi-view consensus and utilize it guide process. Specifically, SCMRL consists within-view reconstruction module module. These modules are elegantly integrated using contrastive strategy, serves align labels both view-specific representations learned simultaneously. This integration allows effectively leverage space, thereby constraining representation. Compared with several state-of-the-art algorithms, extensive experiments demonstrate its superiority. Our code is released https://github.com/YiyangZhou/SCMRL.