Conceitos essenciais
提案されたREALは、EFCILにおける表現強化解析学習を通じて、優れた性能を達成します。
Resumo
この論文では、Exemplar-free class-incremental learning(EFCIL)における表現強化解析学習(REAL)の提案とその効果について説明しています。REALは、デュアルストリームベースの事前トレーニング(DS-BPT)と表現強化蒸留(RED)プロセスを組み合わせて、基本知識を向上させます。実験結果は、REALがEFCILの最先端技術を凌駕し、ALベースのCIL方法と協力してパフォーマンスを向上させることを示しています。
1. Introduction
- CIL allows models to acquire knowledge in phases.
- Catastrophic forgetting is a challenge in CIL.
2. Related Works
- Replay-based and EFCIL methods are compared.
3. The Proposed Method
- REAL enhances representations for unseen data categories.
4. Experiments
- REAL outperforms existing EFCIL methods on various datasets.
Data Extraction:
- "Empirical results on various datasets including CIFAR-100, ImageNet-100 and ImageNet-1k, demonstrate that our REAL outperforms the state-of-the-arts in EFCIL, and achieves comparable or even more superior performance compared with the replay-based methods."
Estatísticas
Empirical results on various datasets including CIFAR-100, ImageNet-100 and ImageNet-1k, demonstrate that our REAL outperforms the state-of-the-arts in EFCIL, and achieves comparable or even more superior performance compared with the replay-based methods.