Workshop on Machine Learning Techniques for Software Development and Optimization (MLOpt 2023)

MLOpt Workshop at HiPEAC 2023

Machine learning and AI techniques have shown considerable promise in automating optimization and decision processes that were previously the realm of human experts or classical combinatorial approaches. The MLOpt workshop addresses the usage of ML and AI techniques in the area of the software development and deployment life-cycle in high-performance computing. Submissions are welcome that report on the usage of machine learning techniques in all areas of software development, porting, optimization, and deployment with a focus on high performance and parallel computing platforms.

Topics of interest include, but are not limited to

• AI/ML-assisted code completion and refactoring methods
• Usage of machine learning in code generation and optimization
• ML-assisted auto-tuning
• Data storage and processing format selection via machine learning (e.g., sparse format selection)
• Dynamic algorithm selection and tuning using ML approaches
• ML-guided automated scheduling in runtime systems
• ML-techniques in performance and correctness tools (e.g., automated data analysis)
• Data driven and ML enhanced mathematical models and algorithms improving energy efficiency, robustness and scalability for extreme computational mechanics problems

Program Committee (under construction)

• Kamer Kaya (Sabancı University, Turkey)
• Didem Unat (Koç University, Turkey)
• Xing Cai (Simula Research Laboratory, Norway)
• Johannes Langguth (Simula Research Laboratory, Norway)
• Aleksandar Ilic (INESC-ID, Portugal)
• Leonel Sousa (INESC-ID, Portugal)
• Karl Fuerlinger (LMU Munich, Germany)
• Vissarion Papadopoulos (National Technical University of Athens)
• George Stavroulakis (National Technical University of Athens)
• Benjamin Cumming (Swiss National Supercomputing Centre)
• Triantafyllos Stylianopoulos (University of Cyprus)
• Ilias Hatzakis (Greek Research and Technology Network)
• Ali Jannesari (Iowa State University, USA)

Organization

MLOpt 2023 is a workshop in the context of the EuroHPC projects SparCity and DComEX. The co-organizers of the workshop are Karl Fuerlinger (LMU Munich, Germany) and George Stavroulakis (NTUA, Greece).

Paper Submission

Prospective authors are invited to submit two-page extended abstracts in the IEEE Conference Proceedings format. There will be no formal published proceedings for the workshop, but we plan make the submissions available to workshop participants to foster discussion during the event.

Please submit your extended abstract on Easychair.

Important dates

• Submission deadline: November 15, 2022 November 30, 2022
• Author notification: November 30, 2022 December 14, 2022
• Final papers due: December 15, 2022 December 21, 2022

Workshop Program

The workshop is scheduled to take place on Monday Jan. 16, 2023 as an in-person event in Toulouse, France.

Session 1: 14:00 - 15:30 (25 min + 5min Q&A)

• Efficient Extraction of Sparse Tensor Features
  Eren Yenigul, Tugba Torun and Didem Unat
• Machine Learning Approaches for Sparse Matrix Vector Optimization
  Konstantin Pogorelov, James Trotter, Xing Cai and Johannes Langguth
• On Efficient Deep Learning for Epistasis Detection
  Miguel Graça, Diogo Marques, Sergio Santander-Jiménez, Leonel Sousa and Aleksandar Ilic

Coffee Break 15:30 - 16:00

Session 2: 16:00 - 17:00 (25 min + 5min Q&A)

• AI-SOLVE - A Machine Learning enhanced library for accelerating the solution of large-scale parametrized systems
  Ioannis Kalogeris, George Stavroulakis and Vissarion Papadopoulos
• From Reactive to Proactive Task Offloading Guided by Machine Learning
  Minh Chung and Karl Fuerlinger