Continual Anomaly Detection on 21 Heterogeneous Tabular Datasets (All)
83.92Balanced AccuracyTaskFusion (AGF+OE+WT+CT+DD)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| TaskFusion (AGF+OE+WT+CT+DD)Replay strategy=Dataset Distillation, Outlier Exposure (OE)=true, WithinTask Augmentation (WT)=true, CrossTask Augmentation (CT)=true2026.06 | 83.92 | 68.85 | 90.54 | |
| AGF+OE+CT+DDReplay strategy=Dataset Distillation, Outlier Exposure (OE)=true, CrossTask Augmentation (CT)=true2026.06 | 83.34 | 66.76 | 89.53 | |
| AGF+OE+WT+DDReplay strategy=Dataset Distillation, Outlier Exposure (OE)=true, WithinTask Augmentation (WT)=true2026.06 | 82.99 | 66.66 | 89.53 | |
| AGF+OE+DDReplay strategy=Dataset Distillation, Outlier Exposure (OE)=true2026.06 | 80.37 | 65.26 | 88.28 | |
| MultitaskContinual Learning protocol (w/o CL vs CL)=Joint2026.06 | 79.73 | 70.09 | 88.88 | |
| Tasks w/o CLContinual Learning protocol (w/o CL vs CL)=w/o CL2026.06 | 78.74 | 68.14 | 86.26 | |
| AGF+DDReplay strategy=Dataset Distillation2026.06 | 75.09 | 62.76 | 87 | |
| AGF+RSReplay strategy=Random Sampling2026.06 | 72.97 | 56.97 | 80.52 | |
| EDSR2026.06 | 67.43 | 48.91 | 71.46 | |
| CaSSLe2026.06 | 63.52 | 46.27 | 69.58 | |
| FinetuneContinual Learning protocol (w/o CL vs CL)=CL2026.06 | 51.12 | 32.88 | 51.5 |