List your company
ResearchFormerly WizWand

Sequential Image Classification on MNIST ordered pixel-by-pixel 1.0 (test)

98.8AccuracyNRNN
6.65630.57854.578.422Feb 9, 2021Mar 21, 2021May 1, 2021Jun 10, 2021Jul 21, 2021Aug 30, 2021Oct 10, 2021
Updated 8mo ago

Evaluation results

MethodLinks
NRNNAdversarial perturbation radius (r)=0.01, Perturbation type=FGSM, Hidden dimension (dh)=128, Training multiplicative noise=0.02, Training additive noise=0.022021.0298.8
NRNNAdversarial perturbation radius (r)=0.01, Perturbation type=FGSM, Hidden dimension (dh)=128, Training multiplicative noise=0.02, Training additive noise=0.052021.0298.8
Lipschitz RNNAdversarial perturbation radius (r)=0.01, Perturbation type=FGSM, Hidden dimension (dh)=1282021.0298.1
coRNNAdversarial perturbation radius (r)=0.01, Perturbation type=FGSM, Hidden dimension (dh)=1282021.0297.5
coRNN# units=128, Number of parameters=34k2021.1096.6
LEM# units=128, Number of parameters=68k2021.1096.6
Lipschitz RNN# units=128, Number of parameters=34k2021.1096.3
expRNN# units=360, Number of parameters=69k2021.1096.2
anti.sym. RNN# units=128, Number of parameters=10k2021.1095.8
NRNNAdversarial perturbation radius (r)=0.05, Perturbation type=FGSM, Hidden dimension (dh)=128, Training multiplicative noise=0.02, Training additive noise=0.052021.0295.5
chrono-LSTM# units=128, Number of parameters=68k2021.1094.6
Exponential RNNAdversarial perturbation radius (r)=0.01, Perturbation type=FGSM, Hidden dimension (dh)=1282021.0294.5
NRNNAdversarial perturbation radius (r)=0.05, Perturbation type=FGSM, Hidden dimension (dh)=128, Training multiplicative noise=0.02, Training additive noise=0.022021.0294.3
GRU# units=256, Number of parameters=201k2021.1094.1
LSTM# units=256, Number of parameters=267k2021.1092.9
NRNNAdversarial perturbation radius (r)=0.1, Perturbation type=FGSM, Hidden dimension (dh)=128, Training multiplicative noise=0.02, Training additive noise=0.052021.0286.8
Lipschitz RNNAdversarial perturbation radius (r)=0.05, Perturbation type=FGSM, Hidden dimension (dh)=1282021.0285.7
coRNNAdversarial perturbation radius (r)=0.05, Perturbation type=FGSM, Hidden dimension (dh)=1282021.0285.5
NRNNAdversarial perturbation radius (r)=0.1, Perturbation type=FGSM, Hidden dimension (dh)=128, Training multiplicative noise=0.02, Training additive noise=0.022021.0279.6
Antisymmetric RNNAdversarial perturbation radius (r)=0.01, Perturbation type=FGSM, Hidden dimension (dh)=1282021.0279.4
NRNNAdversarial perturbation radius (r)=0.15, Perturbation type=FGSM, Hidden dimension (dh)=128, Training multiplicative noise=0.02, Training additive noise=0.052021.0270.6
Exponential RNNAdversarial perturbation radius (r)=0.05, Perturbation type=FGSM, Hidden dimension (dh)=1282021.0259.3
Lipschitz RNNAdversarial perturbation radius (r)=0.1, Perturbation type=FGSM, Hidden dimension (dh)=1282021.0258.9
NRNNAdversarial perturbation radius (r)=0.15, Perturbation type=FGSM, Hidden dimension (dh)=128, Training multiplicative noise=0.02, Training additive noise=0.022021.0258.3
coRNNAdversarial perturbation radius (r)=0.1, Perturbation type=FGSM, Hidden dimension (dh)=1282021.0255.9
Lipschitz RNNAdversarial perturbation radius (r)=0.15, Perturbation type=FGSM, Hidden dimension (dh)=1282021.0237.1
coRNNAdversarial perturbation radius (r)=0.15, Perturbation type=FGSM, Hidden dimension (dh)=1282021.0235.1
Antisymmetric RNNAdversarial perturbation radius (r)=0.05, Perturbation type=FGSM, Hidden dimension (dh)=1282021.0224.7
Exponential RNNAdversarial perturbation radius (r)=0.1, Perturbation type=FGSM, Hidden dimension (dh)=1282021.0219.7
Exponential RNNAdversarial perturbation radius (r)=0.15, Perturbation type=FGSM, Hidden dimension (dh)=1282021.0214.3
Antisymmetric RNNAdversarial perturbation radius (r)=0.1, Perturbation type=FGSM, Hidden dimension (dh)=1282021.0211.4
Antisymmetric RNNAdversarial perturbation radius (r)=0.15, Perturbation type=FGSM, Hidden dimension (dh)=1282021.0210.2