Khái niệm cốt lõi
Replay buffer selection in graph continual learning methods is enhanced by considering both class representativeness and diversity within each class of the replayed nodes, leading to improved model performance.
Tóm tắt
グラフ連続学習におけるリプレイバッファの選択方法を改善するため、再生ノードの各クラス内でのクラス代表性と多様性を考慮することで、モデルのパフォーマンスが向上します。既存の手法に比べて、提案されたカバレッジベースのダイバーシティ(CD)アプローチはよりデータ効率的です。
Thống kê
Existing rehearsal-based GCL methods select the most representative nodes for each class and store them in a replay buffer.
The proposed DSLR model considers both class representativeness and diversity within each class of the replayed nodes.
Extensive experimental results demonstrate the effectiveness of DSLR.