9 abr 1995 ano - 数据集增强 (注入噪声)
Training with noise is equivalent to Tikhonov regularization.
Bishop, C. M.
1995
Descrição:
对于某些模型而言，向输入添加方差极小的噪声等价于对权重施加范数惩罚 (Bishop, 1995a,b)。在一般情况下，注入噪声远比简单地收缩参数强大，特别是噪声被添加到隐藏单元时会更加强
大。向隐藏单元添加噪声是值得单独讨论重要的话题
Adicionado na linha do tempo:
Regularization
Data:
