Low SimCLR basseline

[TsungWeiTsai]

Hi author of CLAE,

Thank you for the insightful paper. It would be great if you plan to release the code in the near future so that I can have a better understanding of your work.

Besides, from your paper, the performance of SimCLR is much lower than the ones reported in the original paper (90+% accuracy on CIFAR-10 training from scratch). I am wondering if there is any reason that you are using a much weaker baseline. Since SimCLR also reported the hyper-parameters they used, it would be interesting to see if CLAE can still improve over the strong baselines. Thank you for your time in advance.