New preprint on regularization properties of noise injection out!
MLP with 1 hidden layer and linear activations on synthetic data. Both injecting noise layer-wise and to all weights lead to minimization of the regularized loss.
This new paper is a follow-up to our Anti-PGD paper (ICML 2022). This time we show that small perturbations induce explicit regularization, which we spell out for a few models. Many new research avenues open up from our Theorem 2.
Hans Kersting
Research Scientist
My research interests lie in machine learning, optimization, and data science.