Executive Summary : | High-performance computing (HPC) machines are increasingly focusing on GPUs, a shift from the traditional CPU-based approach. This shift is driven by the increasing demand for Artificial Intelligence, Machine Learning, and quantum simulation. Sparse tensor computations, crucial for applications like machine learning and computational quantum chemistry, need to adapt to this changing landscape. Traditional techniques for accelerating tensor algebra have been limited by runtime libraries, code generation, and specialized hardware. However, modern HPC machines, which include both CPUs and GPUs, present an opportunity to explore efficient computing techniques. Solutions will be based on efficient data structures, efficient algorithms and code-generation techniques, and scalable parallelization strategies. By addressing these challenges, modern HPC machines can effectively utilize their full potential in applications like machine learning and quantum chemistry. |