Abstract by Hadassah Meyer
Physics and Astronomy
Identifying Sloppy Parameters in Simulated Transmission Loss
In underwater acoustics, predictions of an environment type or the distance to a sound source can be estimated with models that rely on a large set of parameters. Often, some of these parameters are “sloppy,” i.e., the parameters may be removed from the model without changing the prediction. One approach to identifying sloppy parameters is the Manifold Boundary Approximation Method (MBAM). MBAM can be applied to simulated transmission loss models to identify when certain parameters becomes sloppy in an environment. This approach shows at what point a parameter no longer has any information content in the model data and thus cannot be inferred from data. A particularly useful application of this is reducing the run time of a machine learning system by informing the machine when to not search for variables that will not inform the overall result.