Computer Vision News - November 2018

in computer vision. So many things that I can’t even follow! I will not pinpoint a particular paper, but I think there is really interesting work going on in the whole area of self-supervised learning. Either using generative adversarial neural networks or using other techniques, where people are trying to discover high-level concepts in vision – like objects, motion, depth, things like this – without actually exclusively supervising the system. I think this is going somewhere. This is the seed. There is no particular application of this yet, but I think it’s the seed of something very big. The next revolution in computer vision. Probably the next revolution in AI, actually. There’s something I have been talking about in all my talks for the last three years, which is that the future of AI is in self-supervised learning. This way you train a machine to learn about how the world works, without training it for a specific task, but then you can train it for a specific task with very, very little data. This is how humans and animals operate. You started seeing a bunch of papers on this from Facebook, Intel, Google, DeepMind and NVIDIA. There’s this paper on colorization where you train 8 Yann LeCun Computer Vision News Guest “… there is really interesting work going on in the whole area of self-supervised learning!” “The next revolution in computer vision. Probably the next revolution in AI …”