Manifold Regularization for Locally Stable Deep Neural Networks

Charles Jin, Martin Rinard

We apply concepts from manifold regularization to develop new regularization techniques for training locally stable deep neural networks. Our regularizers are based on a sparsifica…