Near-Linear Time Projection onto the ${\ell }_{1,\infty }$ Ball; Application to Sparse Autoencoders

Guillaume Perez, Laurent Condat, Michel Barlaud

Looking for sparsity is nowadays crucial to speed up the training of large-scale neural networks. Projections onto the ${\ell }_{1,2}$ and ${\ell }_{1,\infty }$ are among the most efficie…