Dissertation Talk: Graph-Based Dependency Analysis for Scalable Matrix Factorizations: Parallelism and Communication Avoidance

Event details Date: Tuesday, August 4, 2026 Time: 1:00 PM to 1:00 PM Location: 299 Cory Type: Performing Arts Audience: Faculty,Students About this event Matrix factorizations are fundamental to scientific computing, but scaling them across modern parallel systems is difficult because operations depend on results produced earlier in the computation. This dissertation develops a graph-based framework for understanding and managing these dependencies. It shows how parallelism can be exposed either by tracking dependencies as a factorization evolves or by organizing data and computation so that the required information is available locally in advance. These approaches reduce unnecessary scheduling constraints, communication, and synchronization across multicore processors, GPUs, and distributed-memory systems, providing a common perspective on the design of scalable sparse and dense matrix factorizations. Official event details: https://events.berkeley.edu/eecs/event/324624-dissertation-talk-graph-based-dependency-analysis