{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/neuralpower-predict-and-deploy-energy","title":"NeuralPower: Predict and Deploy Energy-Efficient Convolutional Neural Networks","arxiv_id":"1710.05420","date":"2017-10-15","proceeding":null,"authors":["Ermao Cai","Da-Cheng Juan","Dimitrios Stamoulis","Diana Marculescu"],"abstract":"\"How much energy is consumed for an inference made by a convolutional neural\nnetwork (CNN)?\" With the increased popularity of CNNs deployed on the\nwide-spectrum of platforms (from mobile devices to workstations), the answer to\nthis question has drawn significant attention. From lengthening battery life of\nmobile devices to reducing the energy bill of a datacenter, it is important to\nunderstand the energy efficiency of CNNs during serving for making an\ninference, before actually training the model. In this work, we propose\nNeuralPower: a layer-wise predictive framework based on sparse polynomial\nregression, for predicting the serving energy consumption of a CNN deployed on\nany GPU platform. Given the architecture of a CNN, NeuralPower provides an\naccurate prediction and breakdown for power and runtime across all layers in\nthe whole network, helping machine learners quickly identify the power,\nruntime, or energy bottlenecks. We also propose the \"energy-precision ratio\"\n(EPR) metric to guide machine learners in selecting an energy-efficient CNN\narchitecture that better trades off the energy consumption and prediction\naccuracy. The experimental results show that the prediction accuracy of the\nproposed NeuralPower outperforms the best published model to date, yielding an\nimprovement in accuracy of up to 68.5%. We also assess the accuracy of\npredictions at the network level, by predicting the runtime, power, and energy\nof state-of-the-art CNN architectures, achieving an average accuracy of 88.24%\nin runtime, 88.34% in power, and 97.21% in energy. We comprehensively\ncorroborate the effectiveness of NeuralPower as a powerful framework for\nmachine learners by testing it on different GPU platforms and Deep Learning\nsoftware tools.","url_abs":"http://arxiv.org/abs/1710.05420v1","url_pdf":"http://arxiv.org/pdf/1710.05420v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"neuralpower-predict-and-deploy-energy","repo_url":"https://github.com/caiermao/NeuralPower","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null},{"paper_slug":"neuralpower-predict-and-deploy-energy","repo_url":"https://github.com/cmu-enyac/NeuralPower","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null},{"paper_slug":"neuralpower-predict-and-deploy-energy","repo_url":"https://github.com/enyac-group/neuralpower","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null}],"tasks":[{"task_slug":null,"task_name":"GPU"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1710.05420","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}