Supplement to: Embedding Principle of Loss Landscape of Deep Neural Networks