Towards Physical Plausibility in Neuroevolution Systems

TL;DR

This work addresses the environmental and cost concerns of AI by focusing on inference-power reduction in neuroevolution. It introduces power-aware fitness, a two-output model partitioning approach, and a module reutilization strategy to bias the search toward energy-efficient architectures, with power measured via $pyJoules$ during validation. The primary innovations are the partitioned two-model training within a single run and a mutation scheme that promotes low-power modules, guided by fitness functions that balance accuracy and energy. Results on Fashion-MNIST show up to $29.18$ W ($29.2\%$) power reduction with only a negligible accuracy drop, demonstrating that meaningful energy savings are achievable without sacrificing predictive performance; this has practical implications for deploying neuroevolution-derived models on energy-constrained devices and large-scale deployments.

Abstract

The increasing usage of Artificial Intelligence (AI) models, especially Deep Neural Networks (DNNs), is increasing the power consumption during training and inference, posing environmental concerns and driving the need for more energy-efficient algorithms and hardware solutions. This work addresses the growing energy consumption problem in Machine Learning (ML), particularly during the inference phase. Even a slight reduction in power usage can lead to significant energy savings, benefiting users, companies, and the environment. Our approach focuses on maximizing the accuracy of Artificial Neural Network (ANN) models using a neuroevolutionary framework whilst minimizing their power consumption. To do so, power consumption is considered in the fitness function. We introduce a new mutation strategy that stochastically reintroduces modules of layers, with power-efficient modules having a higher chance of being chosen. We introduce a novel technique that allows training two separate models in a single training step whilst promoting one of them to be more power efficient than the other while maintaining similar accuracy. The results demonstrate a reduction in power consumption of ANN models by up to 29.2% without a significant decrease in predictive performance.

Figures (3)

• Figure 1: Example of a two-output model and its left and right partitions, with the layer marked by the intermediate point in red.
• Figure 2: Evolution of the accuracy over 150 generations.
• Figure 3: Evolution of the power consumption over 150 generations.