Slycat Ensemble Analysis of Electrical Circuit Simulations

With recent advances in computational power, scientists can now run thousands of related simulations to explore a single problem. We refer to such a group of related simulations as an ensemble. More generally, an ensemble can be thought of as a set of samples or observations, each consisting of the same set of variables, in a shared high-dimensional space describing a particular problem domain. Practically, an ensemble is a collection of data sets with common attributes that we wish to analyze as a whole. Thus ensemble analysis is a form of meta-analysis that looks at the combined behaviors and features of the ensemble in an effort to understand and describe the underlying problem space. By looking at the collection as a whole, higher level patterns emerge beyond what can be seen by examining individual simulation runs. As an example, sensitivity analysis is a type of ensemble analysis that evaluates how changes in simulation input parameters correlate with changes in simulation results. In addition to revealing the types and strengths of relationships between inputs and outputs, sensitivity analysis can be used to verify that simulation results are within expected ranges and to validate that the underlying model is behaving correctly. Unexpected results could point to unexplored physical regimes, poorly understood parameter spaces, or something as mundane as flaws in simulation codes. Input parameters form the set of independent variables, and outputs the set of dependent variables. Commonly, sensitivity analyses are performed using either simple regression (correlating a single input to a single output at a time), or multiple regression (correlating a group of inputs to a single output). However, neither of