Could machine learning fuel a reproducibility crisis in science §

• Type:#article
• Date read: 2022-08-17
• Subject: machine learning
• Bibliography: https://www.nature.com/articles/d41586-022-02035-w
• #newsletter

---

Example citation §

Key takeaways §

• The pre-print is Kapoor2022 - Leakage and the Reproducbility Crisis in ML-based Science
• People usually apply ML-models after only a few hours of self-study, and reviewers don’t have time to scrutinize the code.
• Risk of over-optimism related to ML models
• Hype around AI and inadequate checks and balances
• Most important issue: Data leakage

---

Kapoor and Narayanan claim that, once errors are corrected, these models perform no better than standard statistical techniques.

Comparison is logistic regression.