References & Citations
High Energy Physics - Experiment
Title: Application of Transfer Learning to Neutrino Interaction Classification
(Submitted on 7 Jul 2022 (v1), last revised 20 Mar 2023 (this version, v2))
Abstract: Training deep neural networks using simulations typically requires very large numbers of simulated events. This can be a large computational burden and a limitation in the performance of the deep learning algorithm when insufficient numbers of events can be produced. We investigate the use of transfer learning, where a set of simulated images are used to fine tune a model trained on generic image recognition tasks, to the specific use case of neutrino interaction classification in a liquid argon time projection chamber. A ResNet18, pre-trained on photographic images, was fine-tuned using simulated neutrino images and when trained with one hundred thousand training events reached an F1 score of $0.896 \pm 0.002$ compared to $0.836 \pm 0.004$ from a randomly-initialised network trained with the same training sample. The transfer-learned networks also demonstrate lower bias as a function of energy and more balanced performance across different interaction types.
Submission history
From: Andrew Chappell [view email][v1] Thu, 7 Jul 2022 08:02:02 GMT (98kb,D)
[v2] Mon, 20 Mar 2023 10:20:21 GMT (138kb,D)
Link back to: arXiv, form interface, contact.