Predicting Transformer Generalization on MNIST-Transformers 40% threshold
0.898Kendall's tauTransformer-NFN Quasi
Evaluation Results
| Method | Links | |
|---|---|---|
| Transformer-NFN QuasiExtra Parameters (%)=4.54% params ++, Number of runs=52026.04 | 0.898 | |
| Transformer-NFN largeExtra Parameters (%)=57.66% params ++, Number of runs=52026.04 | 0.897 | |
| Transformer-NFNNumber of runs=52026.04 | 0.895 | |
| MLPNumber of runs=52026.04 | 0.874 | |
| XGBoostNumber of runs=52026.04 | 0.869 | |
| STATNNNumber of runs=52026.04 | 0.868 | |
| LightGBMNumber of runs=52026.04 | 0.847 | |
| Random ForestNumber of runs=52026.04 | 0.769 |