哪个权重初始化器最适合模型？ [关闭]

Question

关闭。 这个问题不符合Stack Overflow 准则。 它目前不接受答案。

---

这个问题似乎与帮助中心定义的范围内的编程无关。

2天前关闭。

改进这个问题

model_2 = Sequential()

model_2.add(Conv2D(filters = 4, 
                   activation = 'elu', 
                   kernel_size = (5, 5), 
                   input_shape = (500, 500 ,3), 
                   kernel_initializer = GlorotNormal(seed = 42)))
model_2.add(MaxPooling2D(pool_size = (2, 2)))

model_2.add(Conv2D(filters = 8, 
                   activation = 'elu', 
                   kernel_size = (5, 5), 
                   kernel_initializer = GlorotNormal(seed = 42)))
model_2.add(MaxPooling2D(pool_size = (2, 2)))

model_2.add(Conv2D(filters = 16, 
                   activation = 'elu', 
                   kernel_size = (5, 5), 
                   kernel_initializer = GlorotNormal(seed = 42)))
model_2.add(MaxPooling2D(pool_size = (2, 2)))

model_2.add(Conv2D(filters = 32, 
                   activation = 'elu', 
                   kernel_size = (5, 5), 
                   kernel_initializer = GlorotNormal(seed = 42)))

model_2.add(Flatten())

model_2.add(Dense(50, activation = 'elu'))
model_2.add(Dense(100, activation = 'elu'))
model_2.add(Dense(100, activation = 'elu'))
model_2.add(Dense(50, activation = 'elu'))
model_2.add(Dense(2, activation = 'softmax'))

一般训练模型时最好的权重初始化器是什么，还是因模型而异？

另外，发起人会影响模型的成功吗？