Hi, thank you a lot for your contribution.

I cannot reproduce the results shown in your paper on spleen dataset.
Under 5%, 10% and 50% labeled setting, the final training output does not make sense:

How many epochs do you train to get these results?