Pixel-by-pixel Image Classification on Permuted Sequential MNIST (pMNIST) (test)
98.76AccuracyLSSL
Evaluation Results
| Method | Links | |
|---|---|---|
| LSSLState Transition=Learned2021.10 | 98.76 | |
| LSSL2021.10 | 98.76 | |
| S42021.10 | 98.7 | |
| LSSL-fixedState Transition=Fixed2021.10 | 98.6 | |
| CKConv2021.10 | 98.54 | |
| CKConv2021.10 | 98.54 | |
| LMUFFT2021.10 | 98.49 | |
| LMU-FFT2021.10 | 98.49 | |
| HiPPO-LegS2020.08 | 98.3 | |
| HiPPO-RNN2021.10 | 98.3 | |
| HiPPO-RNN2021.10 | 98.3 | |
| HiPPO-Laguerre2020.08 | 98.24 | |
| TrellisNet2019.10 | 98.13 | |
| TrellisNet2020.08 | 98.13 | |
| TrellisNet2021.10 | 98.13 | |
| TrellisNet2021.10 | 98.13 | |
| HiPPO-LegT2020.08 | 98.03 | |
| Transformer2019.10 | 97.9 | |
| Transformer2020.08 | 97.9 | |
| Transformer2021.10 | 97.9 | |
| Transformer2021.10 | 97.9 | |
| URLSTM + Zoneoutregularization=Zoneout2019.10 | 97.58 | |
| URLSTMregularization=Zoneout2020.08 | 97.58 | |
| LMUimplementation=ours2020.08 | 97.29 | |
| Temporal ConvNet2019.10 | 97.2 | |
| Temporal ConvNet2020.08 | 97.2 | |
| TCN2021.10 | 97.2 | |
| TCN2021.10 | 97.2 | |
| LMUreference=[71]2020.08 | 97.15 | |
| URLSTM2019.10 | 96.96 | |
| URLSTMreference=[31]2020.08 | 96.96 | |
| UR-LSTM2021.10 | 96.96 | |
| URGRU + Zoneoutregularization=Zoneout2019.10 | 96.51 | |
| URGRU2021.10 | 96.51 | |
| UR-GRU2021.10 | 96.51 | |
| LipschitzRNN2021.10 | 96.3 | |
| LipschitzRNN2021.10 | 96.3 | |
| Dilated RNN2020.08 | 96.1 | |
| Dilated RNN2021.10 | 96.1 | |
| IndRNN2019.10 | 96 | |
| IndRNN2020.08 | 96 | |
| IndRNNunits=1282019.11 | 96 | |
| IndRNN2021.10 | 96 | |
| BN-LSTMimplementation=ours, hidden units=1002016.03 | 95.4 | |
| BN-LSTMunits=1002019.11 | 95.4 | |
| LSTM w/f STARNumber of layers=8, units=1282019.11 | 95.4 | |
| r-LSTMTraining Protocol=Full BP2018.03 | 95.2 | |
| r-LSTMlayers=2, auxiliary loss=true2019.10 | 95.2 | |
| r-LSTM2020.08 | 95.2 | |
| r-LSTM2021.10 | 95.2 | |
| r-LSTM2021.10 | 95.2 | |
| LSTMImplementation=ours2019.10 | 95.11 | |
| LSTM2020.08 | 95.11 | |
| LSTM2021.10 | 95.11 | |
| STARNumber of layers=12, units=1282019.11 | 94.7 | |
| Dilated GRU2019.10 | 94.6 | |
| FC URNNunits=5122019.11 | 94.1 | |
| STARNumber of layers=8, units=1282019.11 | 94.1 | |
| sTANH-RNN2016.03 | 94 | |
| STANH-RNNunits=1282019.11 | 94 | |
| GRUNumber of layers=2, units=1282019.11 | 93.9 | |
| AntisymRNNunits=1282019.11 | 93.1 | |
| r-LSTMTraining Protocol=Truncate3002018.03 | 92.8 | |
| p-LSTMTraining Protocol=Full BP2018.03 | 92.8 | |
| p-LSTMTraining Protocol=Truncate3002018.03 | 92.5 | |
| LSTMNumber of layers=2, units=1282019.11 | 91.9 | |
| uRNN2016.03 | 91.4 | |
| uRNN2018.03 | 91.4 | |
| uRNNunits=5122019.11 | 91.4 | |
| Soft orthounits=1282019.11 | 91.4 | |
| LSTMimplementation=ours, hidden units=1002016.03 | 90.2 | |
| RHNNumber of layers=2, units=1282019.11 | 89.5 | |
| LSTMTraining Protocol=Full BP2018.03 | 89.4 | |
| LSTMTraining Protocol=Truncate3002018.03 | 88.8 | |
| iRNN2016.03 | 82 | |
| iRNN2018.03 | 82 | |
| iRNNunits=1002019.11 | 82 | |
| vRNNNumber of layers=1, units=1282019.11 | 44 | |
| TANH-RNN2016.03 | 35 |