From Sparse Data to Scalable Systems: Insights for Emerging Computing Architectures

This presentation introduces the HICREST Lab and its research vision at the intersection of high-performance computing, reliability, and AI systems. It highlights current efforts on two core challenges: improving the efficiency of tile-centric architectures for sparse and irregular data structures, and scaling communication-bound applications across distributed multi-GPU systems. The talk shows how sparse computation, matrix reordering, hierarchical partitioning, and communication-aware design can unlock substantial performance gains while addressing the growing demands of modern AI workloads.