backpropagation

1.关于梯度简单的理解

f(x,y)=xy

可以很容易得到f(x,y)关于x 和y的偏导数
函数关于每个变量的偏导数告诉了你整个函数对于单个变量的敏感程度。

2.链式法则

f(x,y,z)=(x+y)z
可以把上面的公式分解成为
q=x+y 和 f=qz

对y求偏导也同样如此
# set some inputs
x = -2; y = 5; z = -4

写成如下代码

# perform the forward passq = x + y # q becomes 3f = q * z # f becomes -12# perform the backward pass (backpropagation) in reverse order:# first backprop through f = q * zdfdz = q # df/dz = q, so gradient on z becomes 3dfdq = z # df/dq = z, so gradient on q becomes -4# now backprop through q = x + ydfdx = 1.0 * dfdq # dq/dx = 1. And the multiplication here is the chain rule!dfdy = 1.0 * dfdq # dq/dy = 1

2.backpropagation 在sigmoid函数上的求梯度的应用

sigmoid函数如下图所示

想求f对于w 和x 的梯度
下面是除了add mul max 以外最简单的表达式 ，以及相应的导数

下面是sigmoid的导数

下面代码分别对x w求梯度

w = [2,-3,-3] # assume some random weights and datax = [-1, # forward passdot = w[0]*x[0] + w[1]*x[1] + w[2]f = 1.0 / (1 + math.exp(-dot)) # sigmoid funct# backward pass through the neuron (backpropagation)ddot = (1 - f) * f # gradient on dot variable, using the sigmoid gradient derivationdx = [w[0] * ddot, w[1] * ddot] # backprop into xdw = [x[0] * ddot, x[1] * ddot, 1.0 * ddot] # backprop into w# we're done! we have the gradients on the inputs to the circuit

考虑一个更复杂的公式

我们在forward pass 的时候先将这个公式分解成前面部分所阐述的最简单的公式。
代码如下：

x = 3 # example valuesy = -4# forward passsigy = 1.0 / (1 + math.exp(-y)) # sigmoid in numerator   #(1)num = x + sigy # numerator                               #(2)sigx = 1.0 / (1 + math.exp(-x)) # sigmoid in denominator #(3)xpy = x + y                                              #(4)xpysqr = xpy**2                                          #(5)den = sigx + xpysqr # denominator                        #(6)invden = 1.0 / den                                       #(7)f = num * invden # done!                                 #(8)

下面的backpropagation 对forward pass 反向求梯度

# backprop f = num * invdendnum = invden # gradient on numerator                             #(8)dinvden = num                                                     #(8)# backprop invden = 1.0 / den dden = (-1.0 / (den**2)) * dinvden                                #(7)# backprop den = sigx + xpysqrdsigx = (1) * dden                                                #(6)dxpysqr = (1) * dden                                              #(6)# backprop xpysqr = xpy**2dxpy = (2 * xpy) * dxpysqr                                        #(5)# backprop xpy = x + ydx = (1) * dxpy                                                   #(4)dy = (1) * dxpy                                                   #(4)# backprop sigx = 1.0 / (1 + math.exp(-x))dx += ((1 - sigx) * sigx) * dsigx # Notice += !! See notes below  #(3)# backprop num = x + sigydx += (1) * dnum                                                  #(2)dsigy = (1) * dnum                                                #(2)# backprop sigy = 1.0 / (1 + math.exp(-y))dy += ((1 - sigy) * sigy) * dsigy                                 #(1)# done! phew