Image Classification on ImageNet 1K (val) (Accuracy vs Baselines)
80.35Accuracy (%)Unpruned Baseline
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Unpruned BaselineMetric Proxy=Full Capacity, Capacity=100%, Backbone=ResNet-502026.03 | 80.35 | — | — | |
| AGF PruningMetric Proxy=Feature Sensitivity, Capacity=25%, Backbone=ResNet-502026.03 | 64.99 | 0.06 | 0.3 | |
| Random PruningMetric Proxy=Uniform Sampling, Capacity=25%, Backbone=ResNet-502026.03 | 64.93 | — | 0.24 | |
| Wanda PruningMetric Proxy=Weights × Act, Capacity=25%, Backbone=ResNet-502026.03 | 64.69 | -0.24 | — | |
| ℓ1-Norm PruningMetric Proxy=Weights (|W|), Capacity=25%, Backbone=ResNet-502026.03 | 64.54 | -0.39 | -0.15 | |
| Taylor PruningMetric Proxy=Loss Approx (∇W · W), Capacity=25%, Backbone=ResNet-502026.03 | 64.42 | -0.51 | -0.27 |