From Language to Silicon: Programming Systems for Sparse Accelerators

This colloquium talk from the Allen School Colloquia Series features Olivia Hsu from Stanford University presenting "From Language to Silicon: Programming Systems for Sparse Accelerators." Explore how domain-specific accelerator programming systems can be improved through innovative approaches to sparse computation. Learn about the Sparse Abstract Machine (SAM), which introduces a unified abstract machine model and compiler abstraction for sparse dataflow accelerators, and Mosaic, which offers modular and portable compilation solutions for heterogeneous sparse accelerators. Hsu, a final-year Ph.D. candidate at Stanford University and NSF Graduate Research Fellow, discusses the challenges of programming domain-specific accelerators and presents solutions that could democratize access to specialized hardware acceleration. The talk concludes with insights into future directions for programming systems across different accelerator domains.