语义分割，具有稀疏的卷积神经网络，用于微生物中的事件重建

论文标题

语义分割，具有稀疏的卷积神经网络，用于微生物中的事件重建

Semantic Segmentation with a Sparse Convolutional Neural Network for Event Reconstruction in MicroBooNE

论文作者

MicroBooNE collaboration, Abratenko, P., Alrashed, M., An, R., Anthony, J., Asaadi, J., Ashkenazi, A., Balasubramanian, S., Baller, B., Barnes, C., Barr, G., Basque, V., Bathe-Peters, L., Rodrigues, O. Benevides, Berkman, S., Bhanderi, A., Bhat, A., Bishai, M., Blake, A., Bolton, T., Camilleri, L., Caratelli, D., Terrazas, I. Caro, Fernandez, R. Castillo, Cavanna, F., Cerati, G., Chen, Y., Church, E., Cianci, D., Conrad, J. M., Convery, M., Cooper-Troendle, L., Crespo-Anadon, J. I., Del Tutto, M., Dennis, S. R., Devitt, D., Diurba, R., Dorrill, R., Duffy, K., Dytman, S., Eberly, B., Ereditato, A., Evans, J. J., Aguirre, G. A. Fiorentini, Fitzpatrick, R. S., Fleming, B. T., Foppiani, N., Franco, D., Furmanski, A. P., Garcia-Gamez, D., Gardiner, S., Ge, G., Gollapinni, S., Goodwin, O., Gramellini, E., Green, P., Greenlee, H., Gu, W., Guenette, R., Guzowski, P., Hagaman, L., Hall, E., Hamilton, P., Hen, O., Horton-Smith, G. A., Hourlier, A., Itay, R., James, C., de Vries, J. Jan, Ji, X., Jiang, L., Jo, J. H., Johnson, R. A., Jwa, Y. J., Kamp, N., Kaneshige, N., Karagiorgi, G., Ketchum, W., Kirby, B., Kirby, M., Kobilarcik, T., Kreslo, I., LaZur, R., Lepetic, I., Li, K., Li, Y., Littlejohn, B. R., Louis, W. C., Luo, X., Marchionni, A., Mariani, C., Marsden, D., Marshall, J., Martin-Albo, J., Caicedo, D. A. Martinez, Mason, K., Mastbaum, A., McConkey, N., Meddage, V., Mettler, T., Miller, K., Mills, J., Mistry, K., Mohayai, T., Mogan, A., Moon, J., Mooney, M., Moor, A. F., Moore, C. D., Lepin, L. Mora, Mousseau, J., Murphy, M., Naples, D., Navrer-Agasson, A., Neely, R. K., Nienaber, P., Nowak, J., Palamara, O., Paolone, V., Papadopoulou, A., Papavassiliou, V., Pate, S. F., Paudel, A., Pavlovic, Z., Piasetzky, E., Ponce-Pinto, I., Prince, S., Qian, X., Raaf, J. L., Radeka, V., Rafique, A., Reggiani-Guzzo, M., Ren, L., Rochester, L., Rondon, J. Rodriguez, Rogers, H. E., Rosenberg, M., Ross-Lonergan, M., Russell, B., Scanavini, G., Schmitz, D. W., Schukraft, A., Seligman, W., Shaevitz, M. H., Sharankova, R., Sinclair, J., Smith, A., Snider, E. L., Soderberg, M., Soldner-Rembold, S., Soleti, S. R., Spentzouris, P., Spitz, J., Stancari, M., John, J. St., Strauss, T., Sutton, K., Sword-Fehlberg, S., Szelc, A. M., Tagg, N., Tang, W., Terao, K., Thorpe, C., Toups, M., Tsai, Y. -T., Uchida, M. A., Usher, T., Van De Pontseele, W., Viren, B., Weber, M., Wei, H., Williams, Z., Wolbers, S., Wongjirad, T., Wospakrik, M., Wu, W., Yandel, E., Yang, T., Yarbrough, G., Yates, L. E., Zeller, G. P., Zennamo, J., Zhang, C.

论文摘要

我们介绍了语义分割网络Sparsessnet的性能，该网络提供了微酮数据的像素级分类。微酮实验采用液体氩时间投影室来研究中微子性质和相互作用。 Sparsessnet是一个submanifold稀疏卷积神经网络，它提供了基于机器学习的初始算法，该算法在Microboone的$ν_e$ - Apperance振荡分析中提供。该网络经过培训，可以将像素分为五个类，这些类别被重新分类为与当前分析更相关的两个类。 SparsessNet的输出是进一步分析步骤中的关键输入。该技术是在液体氩时间投影室数据中首次使用的，并且与先前使用的卷积神经网络相比，在精度和计算资源利用方面都是一种改进。测试样本上达到的准确性是$ \ geq 99 \％$。对于完整的中微子相互作用模拟，处理一个图像的时间为$ \ $ \ $ 0.5秒，内存使用率为1 GB级别，允许使用最典型的CPU Worker机器。

We present the performance of a semantic segmentation network, SparseSSNet, that provides pixel-level classification of MicroBooNE data. The MicroBooNE experiment employs a liquid argon time projection chamber for the study of neutrino properties and interactions. SparseSSNet is a submanifold sparse convolutional neural network, which provides the initial machine learning based algorithm utilized in one of MicroBooNE's $ν_e$-appearance oscillation analyses. The network is trained to categorize pixels into five classes, which are re-classified into two classes more relevant to the current analysis. The output of SparseSSNet is a key input in further analysis steps. This technique, used for the first time in liquid argon time projection chambers data and is an improvement compared to a previously used convolutional neural network, both in accuracy and computing resource utilization. The accuracy achieved on the test sample is $\geq 99\%$. For full neutrino interaction simulations, the time for processing one image is $\approx$ 0.5 sec, the memory usage is at 1 GB level, which allows utilization of most typical CPU worker machine.

下载PDF全文

下载文献需遵守相关版权规定

论文标题