Can you explain your experience with optimizing deep learning models for deployment on edge devices?