This is a good idea. I think it should be mandatory to publish your code, dataset (or at least a sample of the verification data), and then the trained model w

This is a good idea. I think it should be mandatory to publish your code, dataset (or at least a sample of the verification data), and then the trained model weights if you are publishing a neural network "application" paper. It's very simple to put it on github etc, and then link it in the paper, disappointing when it doesn't happen. This problem doesn't just exist for neural networks. There are lots of papers published presenting some algorithm, but they never provide the source you can quickly check or experiment. Certainly now the technology exists to allow things to be more open (ie github).