Cellular Computational Networks for Sustainable Artificial Intelligence

Mainstream artificial intelligence (AI) implementation overheads are computationally expensive, requiring long training time, huge energy requirements, and are not scalable and sustainable. Development of large-scale neural networks and learning systems such as deep neural networks consume mega-watts of power and burn a lot of energy. This is environmentally detrimental, with high carbon emissions, and will contribute to worsening of environmental and climate conditions. Cellular computational networks (CCNs), a class of sparsely connected dynamic recurrent networks, are known to be scalable and have the potential to reduce computational energy and time overheads of AI implementations without sacrificing AI's performance and accuracy. CCNs are presented in this paper for sustainable AI implementation in this context. This is illustrated with case studies of CCN implementations for handling modern power systems operations and management challenges, especially providing situational intelligence to system operators in energy control centers.