Inplace knowledge distillation with teacher assistant for improved training of flexible deep neural networks | IEEE Conference Publication | IEEE Xplore