Developer Guide for Intel® Data Analytics Acceleration Library 2019 Update 4
The forward two-dimensional (2D) max pooling layer is a form of non-linear downsampling of an input tensor X ∈ R n 1 x n 2 x ... x n p . 2D max pooling partitions the input tensor data into 2D subtensors along dimensions k 1 and k 2, selects an element with the maximal numeric value in each subtensor, and transforms the input tensor to the output tensor Y by replacing each subtensor with its maximum element.
Given:
p-dimensional tensor X ∈ R n 1 x n 2 x ... x n p with input data.
Dimensions k 1 and k 2 along which the kernel is applied
Kernel sizes
m
1 and m2:
where
p
1 and
p
2 are paddings
The problem is to compute the value tensor Y = (y i 1 ...i p ) ∈ R l 1 x ... x l p using the downsampling technique.
The layer computes the value
y
i
1
...i
p
as the maximum element in the subtensor. After the kernel is applied to the subtensor at position
, the index of the maximum
T = (t
i
1
...i
p
) is stored for use by the backward 2D max pooling layer:
s 1 and s 2 are strides
The following figure illustrates the transformation.