Is There a Way to Train Convolutional Networks?

Clip is a method for training visual representations from language. It uses convolutional networks that kicked off this whole deep learning revolution back in 2012 with image data and Alex net. The idea roughly hasn't been around for a long time like Richard you have a paper where you promote way back when we're using contrastive losses to basically bridge images and natural language. And it's incredible how long, how much longer we can push on that.