DUCATI: A Dual-Cache Training System for Graph Neural Networks on Giant Graphs with the GPU.
Xin ZhangYanyan ShenYingxia ShaoLei ChenPublished in: Proc. ACM Manag. Data (2023)
Keyphrases
- neural network
- graph representation
- graph theory
- graph databases
- training process
- graph structure
- weighted graph
- training algorithm
- graph matching
- graph theoretic
- graph mining
- graph construction
- directed graph
- pattern recognition
- graph structures
- structural pattern recognition
- connected components
- subgraph isomorphism
- graph theoretical
- undirected graph
- spanning tree
- graph model
- adjacency matrix
- graph partitioning
- series parallel
- labeled graphs
- graph search
- random graphs
- graph classification
- graph isomorphism
- graph properties
- graph data
- reachability queries
- graph clustering
- feedforward neural networks
- bipartite graph
- multi layer perceptron
- graph kernels
- dynamic graph
- topological information
- adjacency graph
- dense subgraphs
- maximum clique
- minimum spanning tree
- artificial neural networks
- main memory
- frequent subgraphs
- finding the shortest path
- random walk
- graph representations
- real time
- maximum common subgraph
- neural network model
- real world graphs
- prefetching
- small world
- attributed graphs
- neighborhood graph
- parallel processing
- multilayer perceptron
- back propagation
- graph layout
- graph patterns
- massive graphs
- bounded treewidth