AI Energy Efficiency for Edge Deployment

We are entering a new era where embedded devices utilize hardware AI accelerators to transform fields like autonomous vehicles, smart healthcare for the elderly, smart-city infrastructure, and emerging mixed reality and wearable technologies. Specialized neural topologies and ultra-low power hardware are now the focus of attention. These optimizations allow state-of-the-art neural architectures to be implemented in low-power consumer appliances. This enables edge-AI devices to identify their owners, provide privacy-secured access control, and personalize the device’s responses to enhance user experience.

This research project explores the energy balances and trade-offs involved in customizing end-to-end neural architectures, a topic not previously studied in the literature. The project supports a PhD student at TU Dublin in the area of Deployment and Edge AI.