H$^{2}$

Reported on 8 benchmarks across 1 task

Note: results are matched by exact model name. Different papers may use the same name for different model variants.

Methodology8 results

Continual LearningonSplit MNIST (5 tasks)
Top 1 Accuracy %
99.9
Continual LearningonSplit CIFAR-10 (5 tasks)
Top 1 Accuracy %
97.3
Continual LearningonSketch (Fine-grained 6 Tasks)
Accuracy
76.2
best: 80.77 (CondConvContinual)
Continual LearningonStanford Cars (Fine-grained 6 Tasks)
Accuracy
90.6
best: 92.8 (CPG)
Continual LearningonCUBS (Fine-grained 6 Tasks)
Accuracy
84.1
best: 84.26 (CondConvContinual)
Continual LearningonWikiart (Fine-grained 6 Tasks)
Accuracy
75.1
best: 78.32 (CondConvContinual)
Continual LearningonImageNet (Fine-grained 6 Tasks)
Accuracy
75.71
best: 76.16 (ProgressiveNet)
Continual LearningonFlowers (Fine-grained 6 Tasks)
Accuracy
94.9
best: 97.16 (CondConvContinual)