Speaker
Description
I present studies on a deep convolutional autoencoder originally designed to remove electronic noise from a p-type point contact high-purity germanium (HPGe) detector. With their intrinsic purity and excellent energy resolutions, HPGe detectors are suitable for a variety of rare event searches such as neutrinoless double-beta decay, dark matter candidates, and other exotic physics. However, noise from the readout electronics can make identifying events of interest more challenging. At lower energies, where the signal-to-noise ratio is small, distinguishing signals from backgrounds can be particularly difficult.
I demonstrate that a deep convolutional autoencoder can denoise pulses while preserving the underlying pulse shape well. Results show that a deep learning-based approach is more effective than traditional denoising methods. I also present several studies on how the use of this autoencoder can lead to better physics outcomes through improvements in the energy resolution and better background rejection. Finally, I highlight extensions of this research that our group is working on and show how our methods are broadly applicable to the particle astrophysics community.
| Keyword-1 | denoising |
|---|---|
| Keyword-2 | autoencoders |
| Keyword-3 | germanium detectors |
