Empowering Software Development through Machine Learning (ESwML)
April 22nd, 2024
13:45 - 14:30 (incl. 10 min Q&A)
Is Machine Learning Necessary to Use in Cloud Resource Management? Thaleia Dimitra Doudali, IMDEA Software Institute, Madrid, Spain Slides
14:30 - 15:15 (incl. 10 min Q&A)
Towards Transparency in Computational Footprint of Deep Learning Pinar Tözün, IT University of Copenhagen, Denmark Slides
15:45 - 16:30 (incl. 10 min Q&A)
Challenges and Automation When Using Machine Learning Surrogates in Scientific Applications Konstantinos Parasyris, Lawrence Livermore National Laboratory, USA Slides
16:30 - 17:15 (incl. 10 min Q&A)
Auto-HPCnet: an Automatic Framework to Build Neural Network-based Surrogate for HPC Applications Dong Li, University of California, Merced, CA USA Slides
Attendance at this workshop is part of the registration for Eurosys 2024. See here to register.
The software of tomorrow will heavily rely on the use of machine learning models. This will span various aspects including using Machine Learning (ML) models during the development time to enhance developer productivity, designing ML heuristics to improve application execution, and adopting surrogate Neural Networks (NN) models within applications to replace expensive computations and accelerate their performance. However, several challenges limit the broad adoption of ML in today’s software.
For example, there are no programming language extensions that can capture the developer’s intent to use surrogate NN models in their applications, nor can task scheduling algorithms communicate seamlessly with ML heuristics to decide and schedule tasks. As applications continue to get integrated into complex, deep software stacks with workflows, compilers, runtime libraries, and heterogeneous systems, it becomes necessary to use novel techniques for assisting software development, supporting the application execution orchestration, and potentially improving application performance.
The goal of Empowering Software through Machine Learning (ESwML) workshop is to establish a platform where researchers, scientists, application developers, computing center staff, and industry professionals can come together to exchange ideas and explore how artificial intelligence can help in effective and efficient use of future systems.
This workshop will actively drive discussion and aim to answer the following questions:
How can we leverage the advances in Machine Learning to ease the software development process? How can ML be applied to automate heuristic design? These heuristics play a crucial role in: compile-time performance prediction, scheduling policies adopted by datacenters, cloud communities, and HPC programming models. How can we seamlessly integrate ML models into applications to improve their performance while ensuring the correctness of the generated outputs?
Papers must be submitted electronically as PDF files, formatted for 8.5x11-inch paper. The length of the paper must be no more than 8 pages in the ACM double-column format (10-pt font). References are out of the 6 pages limit. Short paper submissions, even with only 2 pages, are welcome as well.
Submissions should use PDF format and be printable on US Letter or A4 paper. Please submit your manuscripts through hotcrp
Proceedings will be posted online. If the final version of an accepted paper does not sufficiently address the comments of the reviewers, then it may be accompanied by a note from the program committee. Publication at ESwML will not prevent later publication in conferences or journals of the presented work. However, simultaneous submission to ESwML and other workshop, conference, or journal is often prohibited by the policy of other venues.
Submission due date: February 15, 2025 (AoE)
Author notification: March 8, 2025
Camera-ready papers: March 13, 2025