Pulse shape identification is essential to filter out unwanted background noise when searching for rare phenomena such as neutrinoless double beta decay or dark matter, especially when using sensitive germanium detectors. Fabrizio Napolitano from the University of Perugia and INFN and his colleagues present a new approach to this challenge by applying quantum machine learning to real experimental data for the first time. The research team developed a hybrid quantum-classical method using variational circuits and succeeded in analyzing germanium detector waveforms with significantly fewer parameters than existing methods. This innovative system achieves surprisingly high levels of accuracy, comparable to the performance of current state-of-the-art methods, while reducing model complexity by more than two orders of magnitude, opening the possibility of streamlined and efficient data processing in future experiments.
Quantum machine learning for neutrino detection
Scientists are pioneering the application of quantum machine learning (QML) to improve event selection in germanium detectors, an important step in the VIP experiment's search for rare phenomena. Researchers investigated variational quantum circuits (VQCs) as classifiers and demonstrated a complete workflow from data preparation to model evaluation, highlighting the potential of QML in this challenging physics application. The team utilized a labeled dataset of waveforms containing examples of both signal and background events from the Broad Energy germanium detector and extracted relevant features to prepare the data for analysis. Performance was evaluated using metrics such as accuracy, precision, and area under the ROC curve, allowing comparisons with traditional machine learning algorithms. The results demonstrate the potential benefits of QML, and the team analyzed the scalability of the approach and evaluated its performance as data size and model complexity increase. This result will contribute to the growing field of quantum computing and its application to fundamental physics research.
Quantum machine learning for pulse shape identification
Scientists have developed a new quantum-classical pipeline for pulse shape identification. This is a technique used to remove background noise in rare event physics experiments utilizing broad energy germanium (BEGe) detectors. This work goes beyond traditional deep learning approaches and pioneers the application of quantum machine learning to real experimental waveforms. The researchers used variational quantum circuits (VQCs) to directly process the 1024-sample waveform and logarithmically compressed the input feature space by mapping it into a Hilbert space of 10 qubits. To establish performance benchmarks, the team referenced state-of-the-art classical pipelines that employ denoising autoencoders (DAEs) and convolutional neural networks (CNNs).
In contrast, quantum pipelines minimize preprocessing and only apply baseline subtraction and normalization, eliminating the need for a dedicated denoising step. VQC's architecture contains only 302 trainable parameters, significantly reducing complexity compared to traditional CNNs. This compact model achieves a remarkable area under the curve of 0.98 and a global accuracy of 97.1%, demonstrating that quantum algorithms can match the performance of established classical baselines, even with current technology. This achievement paves the way for future detectors in which quantum processing units directly analyze the received signal and take advantage of the exponentially large Hilbert space to improve sensitivity.
Quantum machine learning powers rare event detection
Scientists have achieved superior background removal in the search for rare events using a new quantum machine learning approach. The research team developed a 10-qubit variational circuit (VQC) that directly processes the germanium detector's pulse waveform and achieved an area under the curve (AUC) of 0.98 in event classification. The results demonstrate that VQC significantly reduces model complexity by using only 302 trainable parameters, while providing comparable performance to state-of-the-art classical algorithms. As a result of the experiment, the overall accuracy was found to be 97.
1% when evaluating VQC on a test set of 11,377 waveforms. Importantly, the model maintained a high signal efficiency of 98.7% and accurately identified the majority of real events, while achieving 87.7% background rejection. The progressive training strategy employed by the scientists demonstrated a clear correlation between adding quantum layers and improving model performance.
The shallow circuit quickly reached an accuracy of about 94% and captured the overall signature of the pulse. As the circuit depth increases, the accuracy steadily increases, reaching a final value of 97.1%. This shows that the deeper the layers, the better the model can resolve subtle features within the waveform. Benchmarks against classical algorithms revealed model compression ratios of approximately 160x, demonstrating the exceptional parametric efficiency of VQC. This work establishes a path towards compact and efficient signal processing for future physics experiments.
Quantum machine learning improves pulse identification accuracy
This study demonstrates significant advances in pulse shape identification, a key technique for background subtraction in rare event searches using germanium detectors. By applying quantum machine learning to experimental pulse waveforms, the researchers achieved high accuracy with an area under the receiver operating characteristic curve of 0.98 and an overall accuracy of 97.1%. Notably, this performance is comparable to that of established classical deep learning models, but uses a significantly simplified model that requires only 302 trainable parameters.
This achievement avoids the need for computationally intensive preprocessing steps typically required to manage noise in these systems. The team succeeded in the challenge of training deep quantum circuits through a stepwise layer-by-layer training strategy, achieving a high signal acceptance rate of 98.7%, which is essential for sensitive rare event detection. While we acknowledge that this work is a proof of principle, the results suggest the possibility of future detector readout systems in which quantum sensors are directly connected to quantum processing units. This could enable ultra-low-latency, low-power, high-fidelity event classification for next-generation nuclear and particle physics experiments. Future work will explore extending this approach to reconstruct interaction locations and gamma direction by exploiting subtle correlations that are often ignored by traditional filters.
