Implements an efficient softmax approximation as described in the paper "Efficient softmax approximation for GPUs" (http://arxiv.org/abs/1609.04309)

Conditional Focused Neural Question Answering with Large-scale Knowledge Bases