��������regularizer������penalty term�—J(f) � ��������L0�L1 �L2����������������������������������structural risk minimization�SRM������������������������ ��������� ����������������������������������������� �

could work for removing a few weights. However, when pruning a large number of weights, say 60%, we risk the removal of key weights. Hence, a more measured approach to select removal candidates is required

International Conference on Computer Vision. 2019. 9 Zhang, Hongyi, et al. "mixup: Beyond empirical risk minimization." arXiv preprint arXiv:1710.09412 (2017). The cut-mixed sample is calculated as follows:

(4.9.1) 其中l是损失函数，用来度量：给定标签yi，预测f(xi)的“糟糕程度”。统计学家称 (4.9.1)中的这一项为经验 风险。经验风险（empirical risk）是为了近似 真实风险（true risk），整个训练数据上的平均损失，即从其 真实分布p(x, y)中抽取的所有数据的总体损失的期望值： Ep(x,y)[l(f(x), y)] = � � l(f(x), y)p(x set_figsize((4.5, 2.5)) d2l.plot(x, [f(x), g(x)], 'x', 'risk') annotate('min of\nempirical risk', (1.0, -1.2), (0.5, -1.1)) annotate('min of risk', (1.1, -1.05), (0.95, -0.5)) 428 11. 优化算法 11.1.2 深度学习中的优化挑战

products in such equipment or applications and therefore such inclusion and/or use is at customer’s own risk. NVIDIA makes no representation or warranty that products based on this document will be suitable