Pasar al contenido principal

Resultados relacionados

Resultados totales (Incluyendo duplicados): 4
Encontrada(s) 1 página(s)

UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Impact of ML optimization tactics on greener pre-trained ML models

  • González Álvarez, Alexandra|||0009-0003-7634-0343
  • Castaño Fernández, Joel|||0009-0005-6385-9255
  • Franch Gutiérrez, Javier|||0000-0001-9733-8830
  • Martínez Fernández, Silverio Juan|||0000-0001-9928-133X
Machine Learning (ML)-based solutions have currently surpassed human performance in tasks like image classification, visual reasoning, and English understanding. However, this advancement comes at the cost of increasing energy consumption. Traditionally, ML projects have prioritized accuracy over energy, creating a gap in energy consumption during model inference. This study aims to (i) understand image classification datasets and pre-trained models, which is essential for the subsequent analyses, (ii) improve inference efficiency by comparing optimized and non-optimized models, and (iii) assess the economic impact of the optimizations. We conduct a controlled experiment to evaluate the impact of various PyTorch optimization techniques (dynamic quantization, , local pruning, and global pruning) on 42 Hugging Face models for image classification. The metrics examined include GPU utilization, power and energy consumption, accuracy, time, computational complexity, and economic costs. The models are repeatedly evaluated to quantify the effects of these optimization techniques. Dynamic quantization demonstrates inference time and energy consumption reductions, making it suitable for large-scale systems. balances accuracy and energy. Local pruning shows no positive impact on accuracy, and global pruning’s longer optimization times impact economic costs. This study highlights the role of software engineering tactics in achieving greener ML models, offering guidelines for practitioners to make informed decisions on optimization methods that align with sustainability goals., This work has been partially funded by the Spanish research projects GAISSA (TED2021-130923B-I00 by MCIN/ AEI/10.13039/ 501100011033) and DOGO4ML (ref. PID2020-117191RB-I00)., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Contribución de congreso (ConferenceOutput)

Evolution of Kotlin apps in terms of energy consumption: an exploratory study

  • Ahmed, Hesham
  • Boshchenko, Alina
  • Khan, Niaz Ali
  • Knyajev, Dmitriy
  • Garifollina, Dinara
  • Scoccia, Gian Luca
  • Martínez Martínez, Matías-Sebastián|||0000-0002-2945-866X
  • Malavolta, Ivano
© 2023 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes,creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works. https://dx.doi.org/10.1109/ICT4S58814.2023.00014, Context. Java and Kotlin are the two main programming languages used to create Android applications. Kotlin almost completely replicates the capabilities offered by Java and offers extra features, making it a popular choice among developers. From a sustainability perspective, it is crucial to assess the energy usage of Kotlin-based Android applications. Goal. The goal of this study is to explore how the energy consumption of Kotlin applications evolves over time. The study also aims at identifying the key factors that influence energy consumption, to inform developers on how the changes they make affect the energy consumed by their applications. Method. To investigate how Kotlin apps' energy consumption changes through releases, we study three open-source apps Kotlin apps that are also present in the Google Play store. We conduct a measurement-based experiment during which we assess the energy consumed by several releases of each studied application, for a total of 171 executions. Afterwards, we statistically analyse the collected data to identify relevant energy fluctuations (i.e.,, spikes, drops). Finally, we manually inspect the source code changes in the apps to identify possible causes of the identified energy fluctuations. Results. All three studied applications exhibit a growing trend for energy consumption over the course of their releases. Moreover, abnormal energy spikes are found for all applications. There are different causes behind these variations, including OS upgrades, new features, poorly chosen design patterns and libraries, UI issues, and unstable app versions. Conclusions. Our study provides evidence that a number of not fully understood factors can affect the energy consumption of a mobile application. Further work is needed to study their impact., This paper has been funded by the “Ramon y Cajal” Fellowship (RYC2021-031523-I) and the GAISSA Spanish research project (ref. TED2021-130923B-I00; MCIN/AEI/10.13039/501100011033)., Peer Reviewed
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Impact of ML optimization tactics on greener pre-trained ML models

  • González Álvarez, Alexandra
  • Castaño Fernández, Joel
  • Franch Gutiérrez, Javier
  • Martínez Fernández, Silverio Juan
Machine Learning (ML)-based solutions have currently surpassed human performance in tasks like image classification, visual reasoning, and English understanding. However, this advancement comes at the cost of increasing energy consumption. Traditionally, ML projects have prioritized accuracy over energy, creating a gap in energy consumption during model inference. This study aims to (i) understand image classification datasets and pre-trained models, which is essential for the subsequent analyses, (ii) improve inference efficiency by comparing optimized and non-optimized models, and (iii) assess the economic impact of the optimizations. We conduct a controlled experiment to evaluate the impact of various PyTorch optimization techniques (dynamic quantization, , local pruning, and global pruning) on 42 Hugging Face models for image classification. The metrics examined include GPU utilization, power and energy consumption, accuracy, time, computational complexity, and economic costs. The models are repeatedly evaluated to quantify the effects of these optimization techniques. Dynamic quantization demonstrates inference time and energy consumption reductions, making it suitable for large-scale systems. balances accuracy and energy. Local pruning shows no positive impact on accuracy, and global pruning’s longer optimization times impact economic costs. This study highlights the role of software engineering tactics in achieving greener ML models, offering guidelines for practitioners to make informed decisions on optimization methods that align with sustainability goals., This work has been partially funded by the Spanish research projects GAISSA (TED2021-130923B-I00 by MCIN/ AEI/10.13039/ 501100011033) and DOGO4ML (ref. PID2020-117191RB-I00)., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Contribución de congreso (ConferenceOutput)

Evolution of Kotlin apps in terms of energy consumption: an exploratory study

  • Ahmed, Hesham
  • Boshchenko, Alina
  • Khan, Niaz Ali
  • Knyajev, Dmitriy
  • Garifollina, Dinara
  • Scoccia, Gian Luca
  • Martínez Martínez, Matías-Sebastián
  • Malavolta, Ivano
© 2023 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes,creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works. https://dx.doi.org/10.1109/ICT4S58814.2023.00014, Context. Java and Kotlin are the two main programming languages used to create Android applications. Kotlin almost completely replicates the capabilities offered by Java and offers extra features, making it a popular choice among developers. From a sustainability perspective, it is crucial to assess the energy usage of Kotlin-based Android applications. Goal. The goal of this study is to explore how the energy consumption of Kotlin applications evolves over time. The study also aims at identifying the key factors that influence energy consumption, to inform developers on how the changes they make affect the energy consumed by their applications. Method. To investigate how Kotlin apps' energy consumption changes through releases, we study three open-source apps Kotlin apps that are also present in the Google Play store. We conduct a measurement-based experiment during which we assess the energy consumed by several releases of each studied application, for a total of 171 executions. Afterwards, we statistically analyse the collected data to identify relevant energy fluctuations (i.e.,, spikes, drops). Finally, we manually inspect the source code changes in the apps to identify possible causes of the identified energy fluctuations. Results. All three studied applications exhibit a growing trend for energy consumption over the course of their releases. Moreover, abnormal energy spikes are found for all applications. There are different causes behind these variations, including OS upgrades, new features, poorly chosen design patterns and libraries, UI issues, and unstable app versions. Conclusions. Our study provides evidence that a number of not fully understood factors can affect the energy consumption of a mobile application. Further work is needed to study their impact., This paper has been funded by the “Ramon y Cajal” Fellowship (RYC2021-031523-I) and the GAISSA Spanish research project (ref. TED2021-130923B-I00; MCIN/AEI/10.13039/501100011033)., Peer Reviewed, Postprint (author's final draft)