Apart from the content of this research, to me, this method of publishing is the future : here, in one github repo, you can find the paper and the code such that you can perform the experiments done in the research yourself. Well done!
And in a manner of contrast, Google DeepMind’s “Neural Turing Machines” still gives me nightmares. Where are the implementation details, where?! A perfect example of how science should _not_ be reported.
But the most concerning issue however, in my opinion, and taking into account the amount of attention that the NTM paper got, is that (to my knowledge) all public implementations of the suggested architecture involve a great deal of guesswork. The researchers didn’t release any code; only vague details about the architecture and results that we must take on their word. The note they released is, I think, closer to being a tantalizing press release than a proper scientific report.
