高级检索
庞龙刚, 周凯, 王新年. 深度学习在核物理中的应用[J]. 原子核物理评论, 2020, 37(3): 720-726. DOI: 10.11804/NuclPhysRev.37.2019CNPC41
引用本文: 庞龙刚, 周凯, 王新年. 深度学习在核物理中的应用[J]. 原子核物理评论, 2020, 37(3): 720-726. DOI: 10.11804/NuclPhysRev.37.2019CNPC41
Longgang PANG, Kai ZHOU, Xinnian WANG. Deep Learning for Nuclear Physics[J]. Nuclear Physics Review, 2020, 37(3): 720-726. DOI: 10.11804/NuclPhysRev.37.2019CNPC41
Citation: Longgang PANG, Kai ZHOU, Xinnian WANG. Deep Learning for Nuclear Physics[J]. Nuclear Physics Review, 2020, 37(3): 720-726. DOI: 10.11804/NuclPhysRev.37.2019CNPC41

深度学习在核物理中的应用

Deep Learning for Nuclear Physics

  • 摘要: 深度学习是目前最好的模式识别工具,预期会在核物理领域帮助科学家从大量复杂数据中寻找与某些物理最相关的特征。本文综述了深度学习技术的分类,不同数据结构对应的最优神经网络架构,黑盒模型的可解释性与预测结果的不确定性。介绍了深度学习在核物质状态方程、核结构、原子核质量、衰变与裂变方面的应用,并展示如何训练神经网络预测原子核质量。结果发现使用实验数据训练的神经网络模型对未参与训练的实验数据拥有良好的预测能力。基于已有的实验数据外推,神经网络对丰中子的轻原子核质量预测结果与宏观微观液滴模型有较大偏离。此区域可能存在未被宏观微观液滴模型包含的新物理,需要进一步的实验数据验证。

     

    Abstract: Deep learning is the state-of-the-art pattern recognition method. It is expected to help scientists to discover most relevant features from big amount of complex data. Different categories of deep learning, the best deep neural network architectures for different data structures, the interpretability of black-box models and the uncertainties of model predictions are reviewed in this article. The applications of deep learning in nuclear equation of state, nuclear structure, mass, decay and fissions are also introduced. In the end, a simple neural network is trained to predict the mass of nucleus. We found that the artificial neural network trained on experimental data has low prediction error for experimental data that are held back. Trained with experimental data, the network predictions for light neutron-rich nuclei deviate from Macro-Micro Liquid model, which indicate that there might be new physics missing in the theoretical model and more data are needed to verify this.

     

/

返回文章
返回