Double-Exponential Increases in Inference Energy: The Cost of the Race\n for Accuracy

2024/12/12 by Zeyu Yang, Yang, Zeyu, Karel Adámek +3 · 2 citations
Computer Science · #Explainable Artificial Intelligence (XAI)

paper · pdf · doi:10.48550/arxiv.2412.09731

Abstract

Deep learning models in computer vision have achieved significant success but\npose increasing concerns about energy consumption and sustainability. Despite\nthese concerns, there is a lack of comprehensive understanding of their energy\nefficiency during inference. In this study, we conduct a comprehensive analysis\nof the inference energy consumption of 1,200 ImageNet classification models -\nthe largest evaluation of its kind to date. Our findings reveal a steep\ndiminishing return in accuracy gains relative to the increase in energy usage,\nhighlighting sustainability concerns in the pursuit of marginal improvements.\nWe identify key factors contributing to energy consumption and demonstrate\nmethods to improve energy efficiency. To promote more sustainable AI practices,\nwe introduce an energy efficiency scoring system and develop an interactive web\napplication that allows users to compare models based on accuracy and energy\nconsumption. By providing extensive empirical data and practical tools, we aim\nto facilitate informed decision-making and encourage collaborative efforts in\ndeveloping energy-efficient AI technologies.\n

Cited by

• CascadedViT: Cascaded Chunk-FeedForward and Cascaded Group Attention Vision Transformer
• Hardware optimization on Android for inference of AI models

Related

• The Hidden Cost of an Image: Quantifying the Energy Consumption of AI Image Generation
• Watt For What: Rethinking Deep Learning's Energy-Performance Relationship
• Phoeni6: a Systematic Approach for Evaluating the Energy Consumption of Neural Networks
• Full-Cycle Energy Consumption Benchmark for Low-Carbon Computer Vision
• Accuracy is not the only Metric that matters: Estimating the Energy Consumption of Deep Learning Models