The VIP collaboration, a group of researchers from various institutions, has made significant strides in enhancing the sensitivity of the Broad Energy Germanium (BEGe) detector at the Gran Sasso National Laboratory. This detector is crucial for measuring low-energy radiation, particularly in the search for spontaneous collapse-induced radiation and atomic transitions that defy the Pauli Exclusion Principle. The team has introduced a machine learning-based upgrade to the BEGe detector, focusing on event selection to improve the detection of low-energy events down to 10 keV.
The new method employs a denoising autoencoder to suppress electronic and microphonic noises and reconstruct pulse shapes. This step is followed by a convolutional neural network that classifies waveforms as either normal single-site events or those with anomalies. The workflow was rigorously validated on a dataset comprising over 20,000 waveforms recorded in 2021. The classifier achieved impressive results, with a receiver operating characteristic curve area under the curve of 0.99 and an accuracy of 95 percent.
The application of this machine learning procedure has significantly lowered the minimum detectable energy of the final spectrum to approximately 10 keV. This advancement not only enhances the spectral quality but also improves the signal-to-background ratio by about 14 percent and reduces the energy resolution for the characteristic Pb and Bi gamma lines. These improvements are pivotal for increasing the sensitivity of the BEGe detector to rare low-energy signals.
The researchers involved in this project include Simone Manti, Jason Yip, Massimiliano Bazzi, Nicola Bortolotti, Mario Bragadireanu, Ivan Carnevali, Alberto Clozza, Luca De Paolis, Raffaele Del Grande, Carlo Guaraldo, Mihai Antoniu Iliescu, Matthias Laubenstein, Johan Marton, Fabrizio Napolitano, Federico Nola, Kristian Pischicchia, Alessio Porcelli, Alessandro Scordo, Francesco Sgaramella, Diana Sirghi, Florin Sirghi, Johann Zmeskal, and Catalina Curceanu. Their collaborative efforts have provided a scalable framework for future precision tests of quantum foundations in low-background environments.
This research marks a significant leap forward in the field of low-energy radiation detection. By leveraging machine learning techniques, the VIP collaboration has demonstrated the potential to enhance the capabilities of existing detectors, paving the way for more precise and sensitive measurements. The implications of this work extend beyond the immediate scope of the VIP experiment, offering a blueprint for future advancements in quantum physics and related fields.
