Feedforward neural networks前馈神经网络

Feedforward neural networks

or deep feedforward networks

or multilayer perceptrons

Pass input through a series of intermediate computations (hidden layers) to capture non-linear relationships (a.k.a. deep learning)

通过一系列中间计算（隐藏层）传递输入以捕获非线性关系

Train with SGD and Backpropagation（反向传播）(for computing the gradients)

 

NLP applications: word vectors and text classification

A feedforward network : y = f (x; w)

compose together many different functions connected in a chain: f (x) = f3(f2(f1(x)))

 

 embedding layer这一层用来降维

 

Dropout:我们在前向传播的时候，让某个神经元的激活值以一定的概率p停止工作，这样可以使模型泛化性更强，因为它不会太依赖某些局部的特征