Queen's 2015 Nobel Prize in PhysicsTianai Ye has successfully defended her PhD defence. Her thesis is on "Neural Network Methods For Improving Signal Processing In High-purity Germanium Detectors For Rare Event Searches". She carried on her research under the supervision of Prof. Ryan Martin. Congratulations, Dr. Tianai Ye, for the successful defence!
This thesis presents a hybrid convolutional Transformer-autoencoder (CTAE) for self-supervised denoising of high-purity germanium p-type point contact detector signals, trained via the Noise2Noise method on pairs of noisy real data without requiring clean simulated targets. Outperforming a baseline convolutional autoencoder, the model achieves better denoising, signal reconstruction, energy resolution, and a lower effective energy threshold. Beyond denoising, the pretrained encoder is repurposed through transfer learning for drift time estimation — improving low-energy measurements and enabling a full charge trapping correction pipeline — while its latent representation supports unsupervised pulse-shape clustering that resolves finer waveform distinctions than clustering on raw data, with applications to data cleaning in low-energy analyses. Though developed for germanium detectors in the context of neutrinoless double-beta decay experiments, the approach generalizes to other detector technologies, rare event searches, and any domain involving one-dimensional waveforms with paired noisy observations.
The plots below show the results of the denoised waveform evaluated by the machine learning algroithms. The hybrid convulational Transformer-autoencoder result is shown in CTAE-A and CTAE-B.
