Why the lack of reproducibility is crippling research in data mining and what you can do about it

In this talk I will make a strong and potentially controversial claim. The majority of papers published in the best data mining conferences make no contribution. The reason for this is that in most cases, no one, including the original authors can reproduce the findings in the papers. As I shall argue, non-reproducible results are the same as no results at all. The irreproducibility of results may be explicit, the refusal to share data or to give parameter settings, or implicit, the effort to reproduce may be so great that the authors ensure that no one will ever try. I will argue that this lack of reproducibility is crippling research progress, and allowing a large number of false research findings go unchallenged and enter the popular consciousness as true. I will demonstrate my claims with the deconstruction of several influential papers and (reproducible!) experiments.