1 paper · 1 filter
Tiantian Liu, Meng Wan, Jue Wang +1
The start of deep neural network training is characterized by a brief yet critical phase that lasts from the beginning of the training until the accuracy reaches approximately 50\%…