Zoom Talk: Timo Freiesleben (MCMP) und Christoph Molnar (LMU München)
Please contact firstname.lastname@example.org for the password.
Embrace the Complexity: The Paradigm Shift in Science From Statistics to Machine Learning
Machine Learning (ML) is increasingly used in Science, both as a practical tool and as a novel way to gain insight into real-world processes. In some cases, ML simply replaces standard statistical methods; in others, it allows us to address problems that have gone beyond the scope of statistical analysis. We argue that in many Sciences we can observe a methodological paradigm shift from statistical modeling to ML. We discuss two aspects of this shift: (1) A shift towards predictive modeling, which defines a good model as one that generalizes to new data; (2) A shift in explanatory modeling from assumption-based models with interpretable parameters to data-driven models with interpretable model-external descriptions. Finally, we point out problems and limitations that slow or hinder the shift to ML.