具有Sigmoid函数的机器学习算法将其参数初始化为零

我怀疑,这听起来可能很愚蠢,但是我需要深刻地理解这一点。

For a machine learning algo. using numpy and has sigmoid fn. (s = 1/(1+np.exp(-z))

当我们使用它时,我们用零初始化参数。

w = np.zeros((dim,1))
b = 0

So when we are executing z = np.dot(w.T, X) + b As this being a dot product, z must always be zero. I know this is not right. But looking at the prob as w is an array of zero and when it's multiplied with x the result will be zero. This is my doubt. Can someone explain this to me. I appreciate your efforts. Thanks.