GROW: A Row-Stationary Sparse-Dense GEMM Accelerator for Memory-Efficient Graph Convolutional Neural Networks.
Ranggi HwangMinhoo KangJiwon LeeDongyun KamYoungjoo LeeMinsoo RhuPublished in: HPCA (2023)
Keyphrases
- memory efficient
- convolutional neural networks
- graph search
- densely connected
- dense stereo
- dense optical flow
- gaussian graphical models
- iterative deepening
- non stationary
- random walk
- multiple sequence alignment
- external memory
- quasi cliques
- graph structure
- binary matrices
- graph theory
- graph representation
- structured data
- motion field estimation
- convolutional network
- dense sampling
- integral image
- directed acyclic graph
- weighted graph
- graph matching
- face recognition
- dense motion estimation
- spanning tree
- parallel implementation
- graph model
- bipartite graph
- directed graph
- state space
- high dimensional