Backpropagation Through Max-Pooling Layer
Introduction
I have once come up with a question “how do we do back propagation through max-pooling layer?”. The short answer is “there is no gradient with respect to non-maximum values”.
Proof
Max-pooling is defined as
$$
y = \max(x_1, x_2, \cdots, x_n)
$$
where $y$ is the output and $x_i$ is the value of the neuron.
Alternatively, we could consider max-pooling layer as an affine layer without bias terms. The weight matrix in this affine layer is not trainable though.
Concretely, for the output $y$ after max-pooling, we have
$$
y = \sum_{i=1}^{n} w_i x_i
$$
where
$$
w_i =
\begin{cases}
1 & \text{if } x_i = \max(x_1, x_2, \cdots, x_n) \\
0 & \text{otherwise}
\end{cases}
$$
The gradient to each neuron is
$$
\begin{aligned}
\frac{\partial y}{\partial x_i} &= w_i \\
&=
\begin{cases}
1 & \text{if } x_i = \max(x_1, x_2, \cdots, x_n) \\
0 & \text{otherwise}
\end{cases}
\end{aligned}
$$
This simply means that the gradient to the neuron is 1 for the neuron with the maximum value. The gradients for all the other neurons are 0. The neurons whose gradients are 0 do not contribute to the gradients in the earlier neurons due to the chain rule.
Conclusions
There is no gradient with respect to non-maximum values.
Backpropagation Through Max-Pooling Layer