Self Supervised Learning

[h-0-0]

Hi, I've been using avalanche to try out some self-supervised learning in a CL scenario, thought perhaps it may be useful to have some plugins and losses for self-supervised learning in the package.

Currently I have written some code for the NT-Xent loss and a plugin for SimCLR style self supervised training aswell as a SimCLRModel class. They don't employ a CL technique, but could still be useful to have the general framework there to test out CL techniques with SSL?

Could also try writing something more general for SSL than something specifically for SimCLR, but not sure what the best way of approaching that would be.