How neural networks learn one neuron at a time

Original: Mildly Overparameterized ReLU Networks on Orthogonal Data: Incremental Learning and Implicit Bias

Source: arxiv.org ↗