Table 4. Efficient operation of some accelerators which contains data preprocess and efficient architecture.

Name	Data preprocess	Efficient architecture
AWB-GCN (Geng et al., 2020)	CSC	Two TDQs for different sparsity
LW-GCN (Tao et al., 2021)	PCOO	Data replication and row grouping
SPA-GCN (Sohrabizadeh, Chi & Cong, 2022)	Prune the zeros on-the-fly	Input by column-wise and four levels of parallelization
FP-GNN (Tian et al., 2022)	CSR	Outer product and mixed execution
I-GCN (Geng et al., 2021b)	Islandization	Remove redundancy of aggregation