Optimizing Information-theoretical Generalization Bounds via Anisotropic Noise in SGLD

Open in new window