Large language models (LLMs) have shown remarkable potential in various domains, particularly in code optimization and generation for machine programming. However, their performance in low-resource scenarios, such as parallel code generation, remains limited. In this talk, I will discuss our recent advancements in enhancing LLMs’ capabilities for these challenges through agent-based approaches, fine-tuning techniques, and strategic data curation. Additionally, I will introduce our work on representation learning for code data, focusing on tasks like compiler optimization. I will also demonstrate the use of machine learning methods for efficient workload partitioning in system optimization, which can also facilitate scalable distributed training and inference of large-scale machine learning models. Finally, I will share insights into future research directions in these areas. Speaker(s): Guixiang Virtual: https://events.vtools.ieee.org/m/473029