Image Classification on FMNIST (test) (Accuracy, NLL, ECE)
81.4AccuracyDIMS
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DIMSHessian approximation type=low-rank, Backbone=LeNet, Number of parameter samples=100, friction coefficient η0=0.12026.05 | 81.4 | 1.443 | 0.033 | 14.9 | 41.4 | |
| RLAHessian approximation type=low-rank, Backbone=LeNet, Number of parameter samples=1002026.05 | 76.7 | 9.21 | 0.045 | 22 | 52 | |
| LINLAHessian approximation type=low-rank, Backbone=LeNet, Number of parameter samples=1002026.05 | 38.5 | 11.636 | 0.115 | 55.5 | 64.7 | |
| LAHessian approximation type=low-rank, Backbone=LeNet, Number of parameter samples=1002026.05 | 14.7 | 53.519 | 0.167 | 82.5 | 86.6 |