Yann LeCun on his Start in Research and Self-Supervised Learning
The Gradient: Perspectives on AI
00:00
KEMIS Clustering for Joint Embedding Architectures
The best systems are in the low 80% correct, and SWAV is pretty much on top of the heap for that. So basically you train a neural mat in such a way that every training sample is its own category. And they share their weight for the first one, so you can do this iteratively and then kind of work. It's called either quantization or distillation self-supervised learning methods for joint embedding architectures.
Transcript
Play full episode
