Sustainable Deep Learning

Sustainable deep learning focuses on mitigating the substantial environmental and computational costs associated with training and deploying large-scale neural networks. Current research emphasizes developing more efficient model architectures, such as those employing quadratic neurons or shift operations, and optimizing training processes through techniques like hyperparameter optimization and data reduction methods (including coreset selection). These efforts aim to reduce energy consumption and carbon emissions while maintaining or improving model accuracy, thereby promoting the responsible and environmentally conscious development of artificial intelligence.