Appendix for Self-Weighted Contrastive Learning among Multiple Views for Mitigating Representation Degeneration

We denote the hidden layers' features in encoders as Minimizing the left part of Eq. (22) is Eq. For all used datasets, the learning rate is fixed to 0.0003 and the The noise of denoising autoencoder is random Gaussian noise. We do not report the results on Y outubeVideo as this large-scale dataset is beyond the usable range of SVM.