Session

Description Software plays a critical role in scientific discovery across computational science domains, including chemistry, climate science, physics, and applied mathematics. As software development advances, Machine Learning (ML) is becoming an essential tool for enhancing developer productivity, optimizing application execution, and replacing computationally expensive simulations with surrogate Neural Network models. However, several challenges hinder the broad adoption of ML in software, particularly in the context of sustainable development. With increasingly complex software stacks, workflows, and heterogeneous systems, novel techniques are needed to support development, execution orchestration, and performance optimization. A promising approach for reducing software development overhead in High Performance Computing (HPC) is program synthesis, where software is automatically generated from high-level specifications. Large Language Models (LLMs) such as GPT-4, Code Llama, and StarCoder provide intelligent code generation capabilities, yet challenges related to correctness, verification, and reliability remain. Understanding these limitations is crucial for improving ML-driven software development. The ML4SW minisymposium serves as a platform for researchers, developers, and industry professionals to explore ML-driven software synthesis, correctness verification, and application optimization. Key discussions will address ML’s role in enhancing software development, ensuring trustworthiness, and integrating ML into real-world applications for sustainable and efficient computing.