Could the author provide more information on the implementation settings of M-ADA on Cifar10-C ?

[txsing]

• The initial learning rate is 0.1 with a linear decay and the number of epochs is 200. What is the decay rate?

• What are the values of hyper parameters K, T_adv, T_min, gamma, beta, lr_max etc.?

• What are the transformers (Crop \ Normalzaiton etc) used in the training phase?

I tried several times following the settings of origin WRN paper use M-ADA code base, but failed to get the expected results/