Problems with 2D XOR dataset experiments

[wusj0723]

We reproduced the experiment on the XOR data set of your paper, and produced two sets of data respectively for experiment. However, the results were not ideal and there were some deviations from the images in your paper. Did you use any techniques not mentioned in the paper when conducting these experiments? Or is there something special about the distribution of the data you're using? We use two kinds of data:

The results are:

We look forward to hearing from you.

[TDeVries]

I don't think the experiment had anything fancy. It was a pretty straightforward implementation of an MLP + confidence branch trained on a 2D dataset. The dataset was sampled from a continuous uniform distribution in range [-1, 1] and then split into classes based on quadrants. Your dataset looks like it has been rounded to integer values, although I don't think that should make much difference.

How well does your trained model separate the two classes? If the main model is not learning properly then the confidence estimates will also not be great.