Intermittent Inference with Nonuniformly Compressed Multi-Exit Neural Network for Energy Harvesting Powered Devices
This work addresses the challenge of indefinite wait times for inference on energy-harvesting powered devices, which is incremental as it builds on multi-exit networks and compression methods.
The paper tackled the problem of enabling persistent inference on energy-harvesting devices by developing a compression algorithm for multi-exit neural networks that adapts to available energy, resulting in superior accuracy and latency compared to state-of-the-art techniques.
This work aims to enable persistent, event-driven sensing and decision capabilities for energy-harvesting (EH)-powered devices by deploying lightweight DNNs onto EH-powered devices. However, harvested energy is usually weak and unpredictable and even lightweight DNNs take multiple power cycles to finish one inference. To eliminate the indefinite long wait to accumulate energy for one inference and to optimize the accuracy, we developed a power trace-aware and exit-guided network compression algorithm to compress and deploy multi-exit neural networks to EH-powered microcontrollers (MCUs) and select exits during execution according to available energy. The experimental results show superior accuracy and latency compared with state-of-the-art techniques.