Long-range sequence modeling on Sequential MNIST (test)
0.992AccuracyTrellisNet
Evaluation Results
| Method | Links | |
|---|---|---|
| TrellisNetnumber of parameters=8M2018.10 | 0.992 | |
| Dilated GRU2018.10 | 0.99 | |
| IndRNN2018.10 | 0.99 | |
| Generic TCN2018.10 | 0.99 | |
| Transformer (self-attention)2018.10 | 0.989 | |
| r-LSTM w/ Aux. Loss2018.10 | 0.984 |