Energy-Aware Machine Learning Algorithm Design

The exponential increase in machine learning (ML) use on mobile and edge devices indicated a necessity to adopt efficient algorithm design to conserve energy for future consumption and sustainability. Power reduction for energy-constrained platforms like smartphones, Internet of Things devices, and autonomous cars, at training and inference, is critical of importance. This book discusses design techniques for energy-conscious machine learning algorithms, specifically CPU and GPU energy profiling and reducing the power usage with techniques. Profiling techniques and tools are discussed to find out the energy requirements of various algorithms, and model pruning, quantization, knowledge distillation, and low-precision inference are discussed for minimizing inference power usage. For training, efficient backpropagation, energy-conscious optimizers, and distributed training are taken into account. The work also discusses energy efficiency-performance trade-offs and the promise of energy-aware NAS and dynamic resource management. The influence of energy-aware algorithm design is shown through examples of mobile and IoT device, edge computing, and data center applications. Last but not least, hardware constraints and scalability issues are presented, and future directions for designing more energy-efficient ML systems are provided.