Towards Leveraging AutoML for Sustainable Deep Learning: A Multi-Objective HPO Approach on Deep Shift Neural Networks

The article presents a multi-objective hyperparameter optimization (HPO) approach for deep shift neural networks (DSNNs) that aims to enhance model performance while minimizing resource consumption. By integrating state-of-the-art multi-fidelity HPO with multi-objective optimization, the proposed method achieves over 80% accuracy with reduced computational costs. This approach is significant for practitioners as it promotes sustainable AI practices by addressing the environmental impact of deep learning model training and inference.