您的位置: 首页  >  文章  >  Deep learning II - I Practical aspects of deep learning - Dropout regularization 随机失活正则化

Deep learning II - I Practical aspects of deep learning - Dropout regularization 随机失活正则化

分类: 文章 • 2022-10-15 08:28:13

Dropout regularization 随机失活正则化

在训练神经网络时，随机在每层选择一定数量的neuron关闭，从而达到减小过拟合的目的。看起来很疯狂，但是很有效。
Deep learning II - I Practical aspects of deep learning - Dropout regularization 随机失活正则化
Deep learning II - I Practical aspects of deep learning - Dropout regularization 随机失活正则化
Deep learning II - I Practical aspects of deep learning - Dropout regularization 随机失活正则化
1. 对应每一层生成以个mask d[3]
2. 用mask与该层的neurons a[3]覆盖，dropout一些neuron
3. 剩余的neurons除以keep-prob，从而保持该层输出z[3]的期望值不变，避免在测试的时候出现的缩放问题
Deep learning II - I Practical aspects of deep learning - Dropout regularization 随机失活正则化
Deep learning II - I Practical aspects of deep learning - Dropout regularization 随机失活正则化

相关推荐