Ulrich Terstiege

New theory proves SGD convergence in deep linear networks

Researchers have established new theoretical convergence guarantees for stochastic gradient descent (SGD) in the context of deep linear neural networks. This work extends existing gradient descent results to SGD under r…