Perfecting Imperfect Physical Neural Networks with Transferable Robustness using Sharpness-Aware Training.

Control-free and efficient silicon photonic neural networks via hardware-aware training and pruning.