Task-Incremental Learning on Split-CIFAR-100 (20 tasks)
74.8Average AccuracyOGP
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| OGPProtocol=Protocol A, Learning Setting=From-scratch, Backbone=ViT-B/16, Optimizer=Adam2026.04 | 74.8 | 9.1 | 67.2 | |
| ER (5%)Protocol=Protocol A, Learning Setting=From-scratch, Backbone=ViT-B/16, Optimizer=Adam2026.04 | 74.1 | 9.8 | 66.3 | |
| Adam-NSCLProtocol=Protocol A, Learning Setting=From-scratch, Backbone=ViT-B/16, Optimizer=Adam2026.04 | 72.6 | 11.4 | 64.1 | |
| TRGPProtocol=Protocol A, Learning Setting=From-scratch, Backbone=ViT-B/16, Optimizer=Adam2026.04 | 72.4 | 11.6 | 63.8 | |
| SGPProtocol=Protocol A, Learning Setting=From-scratch, Backbone=ViT-B/16, Optimizer=Adam2026.04 | 71.2 | 12.8 | 62.7 | |
| EWCProtocol=Protocol A, Learning Setting=From-scratch, Backbone=ViT-B/16, Optimizer=Adam2026.04 | 64.7 | 21.6 | 51.8 | |
| Sequential FTProtocol=Protocol A, Learning Setting=From-scratch, Backbone=ViT-B/16, Optimizer=Adam2026.04 | 58.3 | 28.4 | 42.1 |