Analysis of Graph Attention Networks and Graph Neural Networks in Node and Graph-Level Tasks. Compared and optimized three different graph based models - Graph Attention Network(GAT), Graph Convolution Network(GCN) and GraphSage.
Results:
- GAT Validation Accuracy - 81%
- GraphSage Validation Accuracy - 79%
- GCN Validation Accuracy - 77%
- Best optimizer for all three was ADAM
Wandb:
Contributors:
- Vishwa Gopalakrishnan
- Aditya Krishna
- Ohm Patel