A Step Toward Quantifying Independently Reproducible Machine Learning Research

What makes a paper independently reproducible? Debates on reproducibility around intuition or assumptions but lack empirical results. Our field on releasing code, which is important, but is not sufficient for reproducibility. We take the first step toward a quantifiable by manually attempting to implement 255 papers published from 1984 until2017, recording features of each paper, and performing statistical analysis of results. For each paper, we did not look at the authors code, if released, order to prevent bias toward discrepancies between code and paper.