Edge AI and TinyML: Optimizing Inference on Low Power Microcontrollers
Keywords:
Edge AI, TinyML, low-power microcontrollers, model quantization, model pruning, inference optimization, energy efficiency, embedded machine learning.Abstract
The rapid advancement of artificial intelligence (AI) has created a growing demand for intelligent systems capable of performing real-time inference with minimal computational resources and energy consumption. Edge AI and Tiny Machine Learning (TinyML) address this challenge by enabling machine learning models to operate directly on resource-constrained devices, such as low-power microcontrollers, without relying on continuous cloud connectivity. This paper explores optimization techniques for deploying and executing machine learning models on microcontrollers with limited memory, processing capability, and power availability. Key approaches include model quantization, pruning, knowledge distillation, operator optimization, and efficient memory management to reduce model size, computational complexity, and inference latency. Furthermore, the study examines the trade-offs between inference accuracy, energy efficiency, execution speed, and memory utilization across TinyML applications, including environmental monitoring, wearable healthcare, industrial automation, and smart sensing.

