Developer Guide for Intel® Data Analytics Acceleration Library 2019 Update 1

Split Backward Layer

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.

Problem Statement

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: