AI_Blog icon indicating copy to clipboard operation
AI_Blog copied to clipboard

反向传播算法的实现,W1计算有个bug

Open paulxin001 opened this issue 4 years ago • 0 comments

Update,W1层的参数时,input = np.sum(batch, axis=0)/batch_size的维度时(3,),这导致input*np.transpose(sig1)的维度是[1,3]而不是[3,3]。更新参数的时候不是按公式来更新。

paulxin001 avatar Jun 15 '21 07:06 paulxin001