Bayesian Neural Networks Avoid Encoding Complex and Perturbation-Sensitive Concepts

Open in new window