Ordered Pixel-by-Pixel Classification on MNIST ordered pixels (test)
99.2AccuracyLipschitz RNN
Evaluation results
| Method | Links | |
|---|---|---|
| Lipschitz RNNtest_perturbation=clean2021.02 | 99.2 | |
| coRNNtest_perturbation=clean2021.02 | 99.1 | |
| NRNNtest_perturbation=clean, training_multiplicative_noise=0.02, training_additive_noise=0.022021.02 | 99.1 | |
| NRNNtest_perturbation=clean, training_multiplicative_noise=0.02, training_additive_noise=0.052021.02 | 99.1 | |
| NRNNtest_perturbation=white noise, sigma=0.1, training_multiplicative_noise=0.02, training_additive_noise=0.022021.02 | 98.9 | |
| NRNNtest_perturbation=white noise, sigma=0.1, training_multiplicative_noise=0.02, training_additive_noise=0.052021.02 | 98.9 | |
| NRNNtest_perturbation=salt and pepper, alpha=0.03, training_multiplicative_noise=0.02, training_additive_noise=0.052021.02 | 98.5 | |
| Lipschitz RNNtest_perturbation=white noise, sigma=0.12021.02 | 98.4 | |
| NRNNtest_perturbation=salt and pepper, alpha=0.03, training_multiplicative_noise=0.02, training_additive_noise=0.022021.02 | 98.3 | |
| Lipschitz RNNtest_perturbation=salt and pepper, alpha=0.032021.02 | 97.6 | |
| Antisymmetric RNNtest_perturbation=clean2021.02 | 97.5 | |
| NRNNtest_perturbation=salt and pepper, alpha=0.05, training_multiplicative_noise=0.02, training_additive_noise=0.052021.02 | 97.1 | |
| Exponential RNNtest_perturbation=clean2021.02 | 96.7 | |
| coRNNtest_perturbation=white noise, sigma=0.12021.02 | 96.6 | |
| coRNNtest_perturbation=salt and pepper, alpha=0.032021.02 | 95.6 | |
| NRNNtest_perturbation=salt and pepper, alpha=0.05, training_multiplicative_noise=0.02, training_additive_noise=0.022021.02 | 95.6 | |
| Lipschitz RNNtest_perturbation=salt and pepper, alpha=0.052021.02 | 93.4 | |
| NRNNtest_perturbation=white noise, sigma=0.2, training_multiplicative_noise=0.02, training_additive_noise=0.052021.02 | 92.2 | |
| NRNNtest_perturbation=white noise, sigma=0.2, training_multiplicative_noise=0.02, training_additive_noise=0.022021.02 | 88.4 | |
| coRNNtest_perturbation=salt and pepper, alpha=0.052021.02 | 88.1 | |
| Exponential RNNtest_perturbation=white noise, sigma=0.12021.02 | 86.7 | |
| NRNNtest_perturbation=salt and pepper, alpha=0.1, training_multiplicative_noise=0.02, training_additive_noise=0.052021.02 | 85.5 | |
| Exponential RNNtest_perturbation=salt and pepper, alpha=0.032021.02 | 83.6 | |
| Lipschitz RNNtest_perturbation=white noise, sigma=0.22021.02 | 78.9 | |
| NRNNtest_perturbation=salt and pepper, alpha=0.1, training_multiplicative_noise=0.02, training_additive_noise=0.022021.02 | 78.7 | |
| Antisymmetric RNNtest_perturbation=salt and pepper, alpha=0.032021.02 | 77.1 | |
| Lipschitz RNNtest_perturbation=salt and pepper, alpha=0.12021.02 | 73.5 | |
| NRNNtest_perturbation=white noise, sigma=0.3, training_multiplicative_noise=0.02, training_additive_noise=0.052021.02 | 73.5 | |
| Exponential RNNtest_perturbation=salt and pepper, alpha=0.052021.02 | 70.7 | |
| Antisymmetric RNNtest_perturbation=salt and pepper, alpha=0.052021.02 | 63.9 | |
| NRNNtest_perturbation=white noise, sigma=0.3, training_multiplicative_noise=0.02, training_additive_noise=0.022021.02 | 62.9 | |
| coRNNtest_perturbation=white noise, sigma=0.22021.02 | 61.9 | |
| coRNNtest_perturbation=salt and pepper, alpha=0.12021.02 | 58.9 | |
| Exponential RNNtest_perturbation=white noise, sigma=0.22021.02 | 58.1 | |
| Lipschitz RNNtest_perturbation=white noise, sigma=0.32021.02 | 47.1 | |
| Antisymmetric RNNtest_perturbation=white noise, sigma=0.12021.02 | 45.7 | |
| Exponential RNNtest_perturbation=salt and pepper, alpha=0.12021.02 | 43.4 | |
| Antisymmetric RNNtest_perturbation=salt and pepper, alpha=0.12021.02 | 42.6 | |
| Exponential RNNtest_perturbation=white noise, sigma=0.32021.02 | 33.3 | |
| coRNNtest_perturbation=white noise, sigma=0.32021.02 | 32.1 | |
| Antisymmetric RNNtest_perturbation=white noise, sigma=0.22021.02 | 22.3 | |
| Antisymmetric RNNtest_perturbation=white noise, sigma=0.32021.02 | 17 |