A Step Toward Quantifying Independently Reproducible Machine Learning Research

Abstract

What makes a paper independently reproducible? Debates on reproducibility center around intuition or assumptions but lack empirical results. Our field focuses on releasing code, which is important, but is not sufficient for determining reproducibility. We take the first step toward a quantifiable answer by manually attempting to implement 255 papers published from 1984 until 2017, recording features of each paper, and performing statistical analysis of the results. For each paper, we did not look at the authors code, if released, in order to prevent bias toward discrepancies between code and paper.

arXiv commentto appear in Proc. Neural Information Processing Systems (NeurIPS), 2019

Cited by in corpus (8)

• Improving Reproducibility in Machine Learning Research (A Report from the NeurIPS 2019 Reproducibility Program)
• Accounting for Variance in Machine Learning Benchmarks
• AI Algorithm for Mode Classification of PCF SPR Sensor Design
• Randomness In Neural Network Training: Characterizing The Impact of Tooling
• A Framework for Cluster and Classifier Evaluation in the Absence of Reference Labels
• Pitfalls in Machine Learning Research: Reexamining the Development Cycle
• Prescriptive and Descriptive Approaches to Machine-Learning Transparency
• Hyperparameter Optimization Is Deceiving Us, and How to Stop It