GNNs crack 150K-node graphs

🔥 What happened Researchers at NTNU dropped FiLSim, a GNN framework that computes graph similarity on graphs with over 150,000 nodes. Existing baselines either run out of memory or fail to learn at that scale. FiLSim scores all test pairs in seconds on a single GPU. 💡 Why it matters Graph Edit Distance and Maximum Common Subgraph are NP-hard, and prior GNN work stayed on toy datasets. The team also ships a sequential graph-editing trick to generate large graphs with reliable GED labels. That unlocks chemistry, program analysis, and social networks at real-world scale. ⚡ Our take Finally someone treats scalability as the actual problem instead of flexing on tiny benchmarks. Whether the simplified pipeline survives noisy real-world graphs is the open question.