Edward Raff, Ph.D., Adjunct Instructor for UMBC’s Graduate Programs in Data Science recently published an intriguing research paper on Quantifying Independently Reproducible Machine Learning.
In this research paper, Raff investigates the concern over whether Artificial Intelligence (AI) and Machine Learning (ML) face a reproducibility issue when it comes to the scientific method.
In this deep dive paper, Raff introduces us to his attempts to implement various ML algorithms from scratch, leading to his ML library called JSAT. He explains what reproducible machine learning is, what makes an ML paper reproducible, and lessons learned from the research.