A5095829508- Parallel Computing and Optimization Techniques
- Stochastic Gradient Optimization Techniques
- Interconnection Networks and Systems
- Advanced Graph Neural Networks
- Advanced Memory and Neural Computing
- Graph Theory and Algorithms
- Advanced Data Storage Technologies
University of Illinois Urbana-Champaign
2024
Sparse matrix dense multiplication (SpMM) is commonly used in applications ranging from scientific computing to graph neural networks. Typically, when SpMM executed a distributed platform, communication costs dominate. Such depend on how scheduled. If it scheduled sparsity-unaware manner, such as with collectives, execution often inefficient due unnecessary data transfers. On the other hand, if fine-grained sparsity-aware communicating only necessary data, can also be high software overhead.
We consider a sparse matrix-matrix multiplication (SpGEMM) setting where one matrix is square and the other tall skinny. This special variant, called TS-SpGEMM, has important applications in multi-source breadth-first search, influence maximization, graph embedding, algebraic multigrid solvers. Unfortunately, popular distributed algorithms like SUMMA deliver suboptimal performance for TS-SpGEMM. To address this limitation, we develop novel distributed-memory algorithm tailored Our approach...