Table 1. Reviewed literature regarding...

Table 1.

Reviewed literature regarding energy and resource usage and related influencing factors

Author	Year	Phases of machine learning services
Author	Year	Pre-training Phase	Training phase	Usage phase	Short description
Desislavov et al.	2023			X	Research about energy consumption and computations trends of deep learning models in the domain of computer vision and NLP
Fischer et al.	2023		X	X	Development of a framework for assessing energy efficiency in ML experiments / tasks
García-Martín et al.	2018		X		Paper describes how to measure energy consumption in different machine learning scenarios
García-Martín et al.	2019		X	X	A review of different possibilities to measure energy consumption in machine learning
Han et al.	2015		X	X	Research describes how energy can be saved due to approaches like pruning, quantization and Huffman coding
Henderson et al.	2020		X	X	A framework for energy and carbon footprints of machine learning was developed
Islam et al.	2023		X		Research on the energy consumption of different machine learning algorithms
Kaack et al.	2022		X	X	Development of a framework for understanding the effects of machine learning on GHG emissions
Lacoste et al.	2019	X*	X	X*	A machine learning emission calculator is developed with focus on training and hardware
Lannelongue et al.	2021	X*	X	X*	A calculator for carbon emission is developed based on e.g. hardware, location and algorithm running aspects
Mavromatis	2024		X	X	Examination of model architectures etc. in training and inference
Patterson et al.	2021	X*	X	X*	Research about the energy usage and carbon emission of large language models
Strubell et al.	2019		X		Research about the energy and environmental costs of NLP training
Wang et al.	2023		X	X	Energy aspects of language model finetuning, pre-training and inference, with the example of Google BERT
Yang et al.	2017			X	Research paper on energy-aware pruning to reduce energy consumption in convolutional neural networks
Zanger et al.	2024		X		Development of a recommender system to compare different machine learning classifiers during training

Author	Year	Phases of machine learning services
Author	Year	Pre-training Phase	Training phase	Usage phase	Short description
Desislavov et al.	2023			X	Research about energy consumption and computations trends of deep learning models in the domain of computer vision and NLP
Fischer et al.	2023		X	X	Development of a framework for assessing energy efficiency in ML experiments / tasks
García-Martín et al.	2018		X		Paper describes how to measure energy consumption in different machine learning scenarios
García-Martín et al.	2019		X	X	A review of different possibilities to measure energy consumption in machine learning
Han et al.	2015		X	X	Research describes how energy can be saved due to approaches like pruning, quantization and Huffman coding
Henderson et al.	2020		X	X	A framework for energy and carbon footprints of machine learning was developed
Islam et al.	2023		X		Research on the energy consumption of different machine learning algorithms
Kaack et al.	2022		X	X	Development of a framework for understanding the effects of machine learning on GHG emissions
Lacoste et al.	2019	X*	X	X*	A machine learning emission calculator is developed with focus on training and hardware
Lannelongue et al.	2021	X*	X	X*	A calculator for carbon emission is developed based on e.g. hardware, location and algorithm running aspects
Mavromatis	2024		X	X	Examination of model architectures etc. in training and inference
Patterson et al.	2021	X*	X	X*	Research about the energy usage and carbon emission of large language models
Strubell et al.	2019		X		Research about the energy and environmental costs of NLP training
Wang et al.	2023		X	X	Energy aspects of language model finetuning, pre-training and inference, with the example of Google BERT
Yang et al.	2017			X	Research paper on energy-aware pruning to reduce energy consumption in convolutional neural networks
Zanger et al.	2024		X		Development of a recommender system to compare different machine learning classifiers during training

Note(s):

*Research does not explicitly focus on the energy consumption in this phase. However, they investigate the energy consumption of the hardware which is used in all phases

Source(s): Created by the authors

[ViewLarge]

Sharing Unavailable