Edge Artificial Intelligence

2Optimizing Deep Learning Models for Edge Devices

Haripriya Saraf¹, Kaavya nair¹, Aditya Kurup¹, Preeti Agarwal²* and Anchit Bijalwan³

¹Artificial Intelligence and Data Science, SVKM’S NMIMS, Maharashtra, India

²School of Technology Management & Engineering, Narsee Monjee Institute of Management Studies (Deemed to-be University), Navi Mumbai, Maharashtra, India

³School of Computing and Innovative Technologies, British University Vietnam, Hanoi, Vietnam

Abstract

Edge intelligence is yielded by combining edge computing along with deep learning (DL), in which processing is relocated closer to data generation at the network edge. A technique such as this eliminates the latency and capacity restrictions encountered in traditional cloud and on-device computing, allowing for real-time intelligent analysis–a vital application area for smart cities, autonomous driving, and video surveillance. It is fueled even more by 5G and IoT, where billions of connected devices provide extremely significant data that needs to be analyzed instantly. Optimizing deep learning models is difficult due to competing objectives such as accuracy against computation, battery consumption, and performance when deployed on edge devices. Techniques such as partitioning, rightsizing, and distributed training improve real-time capabilities while posing new issues in network administration, privacy, and scaling. With the convergence of all of these technologies, edge intelligence should play an important role in ...