CLARA: Transpiler for Cloud built Machine Learning Models into Resource-Scarce Embedded Systems

The increasing aim to use Machine Learning (ML) models to solve many problems targets now the end-devices. Implementing such models in Resource-Scarce Embedded Systems (RSES) improves battery life, privacy, and security. However, due to the lack of resources, the end-devices have called for tools to translate ML models built using Cloud computers to low-level languages with distinct degrees of optimizations. The following paper presents a Transpiler tool to translate trained Machine Learning models to plain C. The tool provides the transpiling of many ML algorithms, pre-processing algorithms, and decomposition models. As the paper shows, compared to other transpiling tools available on the market, CLARA provides the highest number of models to transpile and a multitude of optimization that targets memory and/or processing footprint.