Time: Tuesday, October 27, 2015
2:00pm
MTH1311

Speaker: Professor Michael Robinson (American University)

Title: Examining sonar targets with topological invariants

Abstract: For various reasons, synthetic aperture sonar (SAS) target classification in various clutter contexts is usually done using a data-driven, machine learning approach. Unfortunately, the resulting feature set can be rather inscrutable -- what features is it really using? If you train on simulated data, there is a substantial risk of overfitting or worse. Maybe you've trained on artifacts that won't really happen in actual data! On the other hand, experimental data may contain un-leveraged symmetries that are not adequately captured by simulation. I will describe a principled, foundational analysis of target echo structure through the lens of topological signal processing. I will describe how this can be used to extract of actionable, physics-aware features for classification from simulated and experimental data.