Comparing Attention-based Convolutional and Recurrent Neural Networks: Success and Limitations in Machine Reading Comprehension
This work addresses machine reading comprehension for AI systems, but it is incremental as it builds on existing frameworks and focuses on specific dataset improvements.
The authors tackled machine reading comprehension by proposing a compare-aggregate model with two-staged attention, achieving state-of-the-art results on the MovieQA dataset, and investigated its limitations through adversarial examples and comparisons to human performance.
We propose a machine reading comprehension model based on the compare-aggregate framework with two-staged attention that achieves state-of-the-art results on the MovieQA question answering dataset. To investigate the limitations of our model as well as the behavioral difference between convolutional and recurrent neural networks, we generate adversarial examples to confuse the model and compare to human performance. Furthermore, we assess the generalizability of our model by analyzing its differences to human inference,