Pixel-by-pixel Image Classification on MNIST ordered
99.4AccuracyLipschitz RNN using Euler
Evaluation results
| Method | Links | |
|---|---|---|
| Lipschitz RNN using EulerN=128, # params=≈31K2020.06 | 99.4 | |
| Lipschitz RNN using RK2N=128, # params=≈34K2020.06 | 99.3 | |
| MomentumLSTMN=256, # params=≈270K2020.06 | 99.1 | |
| Lipschitz RNN using RK2N=64, # params=≈9K2020.06 | 99.1 | |
| Lipschitz RNN using EulerN=64, # params=≈9K2020.06 | 99 | |
| Exponential RNNN=360, # params=≈69K2020.06 | 98.4 | |
| Incremental RNNN=128, # params=≈4K/8K2020.06 | 98.1 | |
| Antisymmteric RNNN=128, # params=≈10K2020.06 | 98 | |
| LSTM baselineN=128, # params=≈68K2020.06 | 97.3 | |
| Full Capacity Unitary RNNN=512, # params=≈270K2020.06 | 96.9 | |
| Kronecker RNNN=512, # params=≈11K2020.06 | 96.4 | |
| Unitary RNNN=512, # params=≈9K2020.06 | 95.1 | |
| Sequential NAIS-NetN=128, # params=≈18K2020.06 | 94.3 | |
| Soft orth. RNNN=128, # params=≈18K2020.06 | 94.1 |