Developer Guide for Intel® Data Analytics Acceleration Library 2018 Update 3
The forward split layer copies a tensor from the input argument X ∈ R n 1 x n 2 x ... x n p to multiple output tensors. For more details, see Forward Split Layer.
Given k p-dimensional input gradient tensors G (j) of size n 1 x n 2 x ... x n p , where j=1, ..., k, the backward split layer computes the value: