Towards Understanding Training Dynamics for Mildly Overparametrized Models
Abstract: While over-parameterization is widely believed to be crucial for the success of optimization for the neural networks, most existing theories on over-parameterization do not fully explain the reason — they […]