arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2211.00564physics.ins-detastro-ph.IM
收藏

在 CRESST 中利用深度学习实现自动化数据清洗

Towards an automated data cleaning with deep learning in CRESST

G. Angloher, S. Banik, D. Bartolot, G. Benato, A. Bento, A. Bertolini, R. Breier, C. Bucci, J. Burkhart, L. Canonica, A. D'Addabbo, S. Di Lorenzo, L. Einfalt, A… 展开作者

G. Angloher, S. Banik, D. Bartolot, G. Benato, A. Bento, A. Bertolini, R. Breier, C. Bucci, J. Burkhart, L. Canonica, A. D'Addabbo, S. Di Lorenzo, L. Einfalt, A. Erb, F. v. Feilitzsch, N. Ferreiro Iachellini, S. Fichtinger, D. Fuchs, A. Fuss, A. Garai, V. M. Ghete, S. Gerster, P. Gorla, P. V. Guillaumon, S. Gupta, D. Hauff, M. Ješkovský, J. Jochum, M. Kaznacheeva, A. Kinast, H. Kluck, H. Kraus, M. Lackner, A. Langenkämper, M. Mancuso, L. Marini, L. Meyer, V. Mokina, A. Nilima, M. Olmi, T. Ortmann, C. Pagliarone, L. Pattavina, F. Petricca, W. Potzel, P. Povinec, F. Pröbst, F. Pucci, F. Reindl, D. Rizvanovic, J. Rothe, K. Schäffner, J. Schieck, D. Schmiedmayer, S. Schönert, C. Schwertner, M. Stahlberg, L. Stodolsky, C. Strandhagen, R. Strauss, I. Usherov, F. Wagner, M. Willers, V. Zema, W. Waltenberger

AI总结:

本文将 CRESST 暗物质实验的信号清洗建模为时间序列分类,使用四种神经网络在百万级标注数据上训练,最佳模型平衡准确率达 0.932,验证了深度学习自动化数据清洗的有效性。

摘要:

CRESST 实验采用低温量能器,对暗物质粒子诱发的核反冲进行高灵敏度测量。记录到的信号需要经过细致的清洗流程,以避免由堆积和读出伪迹导致的反冲能量错误重建。我们将这一流程构建为时间序列分类任务,并提出利用神经网络使其自动化。借助 CRESST 在 2013 年至 2019 年间记录的、来自 68 个探测器的超过一百万条带标签记录组成的数据集,我们测试了四种常用神经网络架构学习该数据清洗任务的能力。表现最佳的模型在测试集上取得了 0.932 的平衡准确率。我们在一个示例性探测器上表明,约一半被错误预测的事件实际上是被错误标注的事件,而其余事件中很大一部分具有依赖于上下文的真实标签。我们还利用模拟数据评估了分类器的召回率和选择性。结果证实,训练得到的分类器非常适合该数据清洗任务。

英文摘要:

The CRESST experiment employs cryogenic calorimeters for the sensitive measurement of nuclear recoils induced by dark matter particles. The recorded signals need to undergo a careful cleaning process to avoid wrongly reconstructed recoil energies caused by pile-up and read-out artefacts. We frame this process as a time series classification task and propose to automate it with neural networks. With a data set of over one million labeled records from 68 detectors, recorded between 2013 and 2019 by CRESST, we test the capability of four commonly used neural network architectures to learn the data cleaning task. Our best performing model achieves a balanced accuracy of 0.932 on our test set. We show on an exemplary detector that about half of the wrongly predicted events are in fact wrongly labeled events, and a large share of the remaining ones have a context-dependent ground truth. We furthermore evaluate the recall and selectivity of our classifiers with simulated data. The results confirm that the trained classifiers are well suited for the data cleaning task.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑