Speaker
Layne Bradshaw
Description
Anomaly detection with autoencoders is a popular method to search for new physics in a model-agnostic manner.
In this talk, we try to understand these "black boxes" by designing mimickers with a small number of energy flow polynomials as inputs.
These mimickers perform comparably to the autoencoder when ordering background events, but also match the anomaly detection capabilities of the autoencoder across a variety of signal events.
Thus, this approach allows one to create simple, interpretable anomaly detectors.
Author
Layne Bradshaw