Vision Data Selection on CIFAR-10 1k 10% label noise
56.1Accuracy (20% Selected Data)GRASP
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GRASPBackbone=ResNet18, Pre-trained=ImageNet2026.06 | 56.1 | 69.6 | 71.8 | 76.1 | |
| InRun-1Backbone=ResNet18, Pre-trained=ImageNet2026.06 | 55.6 | 68.9 | 71.9 | 75.5 | |
| InRun-2Backbone=ResNet18, Pre-trained=ImageNet2026.06 | 52.8 | 66.4 | 73.4 | 75.9 | |
| RandomBackbone=ResNet18, Pre-trained=ImageNet2026.06 | 50.9 | 63.2 | 70 | 74.5 | |
| DatamodelsBackbone=ResNet18, Pre-trained=ImageNet2026.06 | 47.6 | 64.1 | 71.4 | 74.1 | |
| InfluenceBackbone=ResNet18, Pre-trained=ImageNet2026.06 | 45.8 | 64.4 | 72.2 | 76.9 | |
| KNN-ShapleyBackbone=ResNet18, Pre-trained=ImageNet2026.06 | 34.1 | 48.7 | 64.7 | 74.3 | |
| TRAKBackbone=ResNet18, Pre-trained=ImageNet2026.06 | 29.9 | 43.4 | 52.8 | 65 |