Data-Driven Parameter Reduction in Sloppy Models

The past decades have seen significant research progress towards developing reduced, low-dimensional descriptions of complex dynamic models, leading to simpler systems that are easier to interpret. However, the parallel field of reducing the number of parameters required to capture relevant system behavior is relatively nascent: the Manifold Boundary Approximation Method and active subspaces are two promising approaches to this problem of model “sloppiness”. In this talk we present advances in our own, data-based technique that is capable of discovering sloppy directions in a high-dimensional parameter space without recourse to the analytical equations that constitute the model. We illustrate its effectiveness in the context of biological networks, and show how these methods can be used to perform selective optimization over important parameter combinations, ignoring sloppy directions.