Tiny Machine Learning for Resource-Constrained Microcontrollers

We use 250 billion microcontrollers daily in electronic devices that are capable of running machine learning models inside them. Unfortunately, most these highly constrained terms computational resources, such as memory usage or clock speed. These exactly the same resources play a key role teaching and model with basic computer. However, microcontroller environment, make critical difference. Therefore, new paradigm known tiny had to be created meet requirements embedded devices. In this review, we discuss resource optimization challenges different methods, quantization, pruning, clustering, can used overcome difficulties. Furthermore, summarize present state frameworks, libraries, development environments, tools. The benchmarking is another thing concerned about; constraints diversity hardware software turn benchmark must resolved before it possible measure performance differences reliably between also emerging techniques approaches boost expand process improve data privacy security. end, form conclusion about its future development.