1 paper
Ning Yang, Yikuan Zhang, Qi Ouyang +2
Stochastic gradient descent (SGD) is central to deep learning, yet the dynamical origin of its preference for flatter, more generalizable solutions remains unclear. Here, by analyz…