Conference paperTrimming the 1 regularizer: Statistical analysis, optimization, and applications to deep learning