
AI Is Learning at the Wrong Level of Abstraction - Matthieu Wyart
Deep networks can recover hierarchical abstractions from data, while predicting learned latent representations may reduce the samples needed to acquire them.
Videos about neural training methods, learned representations, model architectures, and the foundations of modern AI. 1 video.

Deep networks can recover hierarchical abstractions from data, while predicting learned latent representations may reduce the samples needed to acquire them.