Machine learning applications are in constant evolution and have become an integral part of our daily lives. Nevertheless, research in this field has mainly focused on producing high-precision models without taking energy constraints into account. This is mainly the case for deep learning, where the aim has been to produce more accurate model results without any constraints in terms of computing or energy consumption. The implications of climate change and the global consensus of Nations have made this trend no longer tolerable. Energy efficiency management must be an additional constraint in the development of learning models, with the aim of reducing the carbon impact of digital technology. Machine learning models such as deep neural networks are characterized by hyper parameters or weights that are used to transform input data into features. These models consist of two distinct phases: the training phase and the validation phase. In the phase of model definition, we have to specify the number of layers, the size, type of each layer and type of activation function parameters in order to learn our model. The aim of this thesis work is to evaluate models’ quality according to three constraints: accuracy, performance and energy consumption, and then to recommend the model that optimizes all three constraints. To this end, an empirical study of each hyper-parameters is performed to determine its impact on one of the quality constraints.
References:
- Eva García-Martín and al., "Estimation of energy consumption in machine learning" 2019 Journal of Parallel and Distributed Computing 134 (2019) 75–88.
- Radwa Elshawi, and al. DLBench: a comprehensive experimental evaluation of deep learning frameworks. Cluster Computing
- DEMBELE, Simon Pierre, and al. Towards Green Query Processing-Auditing Power Before Deploying. In: 2020 IEEE International Conference on Big Data (Big Data). IEEE, 2020. p. 2492-2501.
- Christian Janiesch and al. Machine learning and deep learning, 2021, Electronic Markets
Contact Information:
Simon Dembele
simon [dot] pierre [dot] dembele [at] ut [dot] ee
Radwa El Shawi
Radwa [dot] elshawi [at] ut [dot] ee