PrototypeNAS: Rapid Design of Deep Neural Networks for Microcontroller Units

PrototypeNAS is a new zero-shot neural architecture search method that rapidly designs and optimizes deep neural networks for microcontroller units without requiring extensive training. The system uses a three-step approach combining structural optimization, ensemble zero-shot proxies, and Hypervolume subset selection to identify efficient models within minutes that can run on resource-constrained edge devices.

• →PrototypeNAS eliminates the need for training multiple DNNs from scratch, significantly reducing time and computational resources required for neural architecture search.
• →The method combines structural optimization across multiple architecture types with pruning and quantization configurations in a unified search space.
• →An ensemble of zero-shot proxies is used during optimization instead of relying on a single proxy for better performance estimation.
• →Hypervolume subset selection distills the most meaningful accuracy-FLOP tradeoffs from Pareto front optimization results.
• →Testing across 12 datasets in image classification, time series classification, and object detection shows the method can identify MCU-deployable models in minutes.