AI.info

Cédric Koller

Explore Cédric Koller on AI.info.

• A solvable high-dimensional model where nonlinear autoencoders learn structure invisible to PCA while test loss misaligns with generalization