Reproducibility via Crowdsourced Reverse Engineering: A Neural Network Case Study With DeepMind's Alpha Zero
share
The reproducibility of scientific findings are an important hallmark of quality and integrity in research. The scientific method requires hypotheses to be subjected to the most crucial tests, and for the results to be consistent across independent trials. Therefore, a publication is expected to provide sufficient information for an objective evaluation of its methods and claims. This is particularly true for research supported by public funds, where transparency of findings are a form of return on public investment. Unfortunately, many publications fall short of this mark for various reasons, including unavoidable ones such as intellectual property protection and national security of the entity creating those findings. This is a particularly important and documented problem in medical research, and in machine learning. Fortunately for those seeking to overcome these difficulties, the internet makes it easier to share experiments, and allows for crowd-sourced reverse engineering. A case study of this capability in neural networks research is presented in this paper. The significant success of reverse-engineering the important accomplishments of DeepMind's Alpha Zero exemplifies the leverage that can be achieved by a concerted effort to reproduce results.
READ FULL TEXT