Weighted Regularization Keeps Sparse Learning Alive Across Tasks

Original: Continual Classification Beyond the Kernel Regime: Harnessing the Sparsity Bias of L2 Regularization in Diagonal Linear Networks

Source: arxiv.org ↗