Authors: Satyender Chauhan , Durgesh Mishra
Abstract: Generative Artificial Intelligence (AI) has rapidly evolved in recent years, enabling machines to produce text, images, audio, and video with high fidelity. Traditionally, these models rely on high-performance cloud servers due to their computational and memory demands. However, with the proliferation of Internet-of-Things (IoT) devices and embedded systems, there is an increasing need to deploy generative AI models directly on edge devices. This review explores the landscape of edge generative AI models on embedded hardware, highlighting challenges, optimization strategies, hardware architectures, and application areas. We examine model compression techniques, hardware accelerators, and energyefficient designs that allow sophisticated AI models to run on resource-constrained devices. We also discuss future directions, including federated learning, privacy-preserving AI, and ultra-low latency inference. The paper concludes with a comprehensive assessment of stateof-the-art practices and recommendations for integrating generative AI in embedded systems.
Keywords: Edge AI, Generative AI, Embedded Hardware, Model Compression, Low-Power AI, On-Device Inference
Full Issue
| View or download the full issue | PDF 68-81 |