Sociomeetings journal club: Sloppy Models. Why (Almost) Every Model Has Too Many Parameters

Note that the sociomeeting will start at 14:00 and will be at the seminar room.

Ever built a model, tweaked a bunch of parameters, and realized that most of them barely matter? Welcome to the world of sloppy models. Many complex systems have models with parameters that seem redundant, yet still manage to make accurate predictions. This insight has gained significant attention within the Spanish complex systems community, as it can rigorously explain how models can remain predictive despite high-dimensional parameter spaces.

In this discussion, we’ll explore how parameter space compression helps us understand emergent theories, why some directions in parameter space are crucial while others are just noise, and what this means for modeling in general. Our guide for the session will be two key papers: a short one in Science for those in a hurry, and a more complete discussion from The Journal of Chemical Physics for those who want to go further.

Attendants are expected to have read at least one of the following papers.

Transtrum, M. K., Machta, B. B., Brown, K. S., Daniels, B. C., Myers, C. R., & Sethna, J. P. (2015). Perspective: Sloppiness and emergent theories in physics, biology, and beyond. The Journal of chemical physics, 143(1). arXiv

 

Machta, Benjamin B., et al. "Parameter space compression underlies emergent theories and predictive models." Science 342.6158 (2013): 604-607. (arXiv)