Describe your approach to fine-tuning a pre-trained vision model for a specific low-power application on edge devices like Syntiant targets.
