Parallelizing GNNs
Schedule
11/03 Create data loaders for graphs (Shane) and utilities for partitioning (Peter).
11/10 Implement sequential and naive parallel implementation (Shane and Peter).
11/17 Test sequential (Peter) and baseline parallel implementation (Shane).
11/21 Optimize partitioning (Peter) and speed up slowest parts of GNN evaluation (Shane).
11/24 Create data loader for variety of graph types (Peter), identify other potential modifications to GNN (Shane), conduct ablation studies (Shane and Peter).
11/28 Use the measurements done on sparse graphs to modify our algorithm & test again (Shane and Peter).
12/01 Make final measurements on random, sparse graphs (Shane), writeup draft (Peter), and demo (Shane and Peter).
The remaining time will serve as a buffer to run additional experiments and revise the writeup (Shane and Peter).
Proposal
Link
Checkpoint
Link
Final
Link