Created by: qingqing01

• ConvexCombinationLayer.cpp

This is really just a special case of batched matrix multiply: https://github.com/PaddlePaddle/Paddle/issues/4683

weight: k x d x 1 (reshaped from k x d) input: k x n x d output: k x n x 1 (batched matrix mult of input x weight)

(we can flatten back to k x n)