
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 how AI systems learn internal features and abstractions from data rather than relying on hand-built rules. 1 video.

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