
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 network architectures, training behavior, learned features, and the computational structures behind AI models. 1 video.

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