IACS Seminar, Tuning models for skill: When is prediction better than accommodation?
Thursday, Sept. 27th, 2018
1:00 - 2:00 pm
IACS Seminar room

In a variety of disciplines, highly idealized simulation models play an important role in gaining understanding and predictive power. Because of the significant role played by idealization, models often require the "tuning" of parameter values in order to accommodate existing data. It is widely understood that when this occurs, testing the models against the very data used to tune them ("double counting") is highly undesirable, and it is much better to be able to test models against novel data. Philosophers of science (as we tend to do) have disagreed about these questions. I present a framework for understanding model tuning and argue that how serious of a problem "double counting" is depends on one's scientific situation.

Bio:
Eric Winsberg is Prof. of Philosophy at the University of South Florida. He is author of Science in the Age of Computer Simulation (Chicago) and Philosophy and Climate Science (Cambridge).
https://calendar.stonybrook.edu/site/iacs/event/tuning-models-for-skill--when-is-prediction-better-than-accommodation/