Full-stack Optimization of Ultra-low-power TPGs for Intelligent Cyberphysical Systems – foutics

The increasing complexity of deep neural network based artificial intelligence (AI) is creating a challenge for embedding AI in ultra-low power (ULP) Internet-of-Things (IoT) devices and Cyber- Physical Systems (CPS). The aim of the FOUTICS project is to propose full-stack methods to train and infer ultra-lightweight AIs, by extending, implementing, and optimizing Tangled Program Graphs (TPGs), a new light-by-construction machine learning technique based on genetic programming principles. Exploiting the TPG efficiency and integrating energy optimization at its core, FOUTICS will create methodologies to implement energy efficient AI, capable of nanoseconds reaction time on hardware platforms ranging from ULP embedded devices to reconfigurable devices. Extending TPG to new learning environments, FOUTICS will enable real world CPS use cases and will optimize energy use from training to physical execution on the factory floor.