MEGA pre-trains a GNN with MAML-style second-order gradients over simulated task sequences plus knowledge distillation and replay, improving graph few-shot class-incremental learning without using query sets during incremental training.
Inductive representation learning on large graphs,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
MEGA: Second-Order Gradient Alignment for Catastrophic Forgetting Mitigation in GFSCIL
MEGA pre-trains a GNN with MAML-style second-order gradients over simulated task sequences plus knowledge distillation and replay, improving graph few-shot class-incremental learning without using query sets during incremental training.