Energyefficient Artificial Intelligence (EEAI, German: Energieeffiziente KI – EEKI) is one of the frontier projects for energy-efficient technology development by the German Energy Agency GmbH dena on behalf of the German Federal Ministry for Economic Affairs and Climate Action (BMWK). The project is part of dena’s Future Energy Lab – the pilot lab for the digitalisation of the energy transformation.
Machine Learning and artificial intelligence have evolved into almost every technical field by providing application-specific neural networks for high precision data classification, analysis, training, inference, object detection, transportation, network optimization, multi-model data understanding, data compression and many more. This was achieved by increasingly complex neural network architectures with millions of neurons, weights and parameters that also require more and more processing power and thus energy consumption for training, inference and transmission – the latter mainly for distributed or federated learning algorithms with continuous neural network update requirements.
Here, EEAI will address this topic by developing methods for optimizing and reducing energy consumption in artificial intelligence applications. In particular, the project will first investigate specific FPGA-based hardware for energy-efficient neural network inference in data centers. And second, optimized transmission scenarios for neural network updates with minimum data exchange and overall energy savings will be investigated, as distributed AI solution become more and more common.