A Convolutional Network Approach to Machine Comprehension

Machine Comprehension is a daunting task, since it requires cross-encoding and exchanging information between a context paragraph and a given query in order to produce an answer span. In designing baselines for a machine comprehension model, each model training has a long turnover, which does not bode well when there is limited time to train. Long runtimes are often from implementing recurrent neural networks (RNNs), whose forward and backward passes do not make adequate use of the parallel compute units at hand. This paper discusses how to apply convolutional neural networks (CNNs) to the machine comprehension task. The author incorporates CNNs with existing bidirectional attention-flow mechanisms and compares the performance to RNN-based models. The model has been evaluated on the Stanford Question Answering Dataset (SQuAD).