A Question-Answering framework for plots using Deep learning

Published in 2019 International Joint Conference on Neural Networks (IJCNN), 2018

Paper

Short Summary

  • Created a novel architecture comprising depth-wise, 1-D convolutions and LSTMs, for question-answering on scientific plots. \item Achieved state-of-the-art accuracy (83.95%) on FigureQA Dataset (Maluuba-Microsoft), bettering Relational Networks (Google DeepMind) by 8.53% and reducing the training time by 93% with 75% lesser computational resources.

Abstract

Deep Learning has managed to push boundaries in a wide variety of tasks. One area of interest is to tackle problems in reasoning and understanding, in an aim to emulate human intelligence. In this work, we describe a deep learning model that addresses the reasoning task of question-answering on bar graphs and pie charts. We introduce a novel architecture that learns to identify various plot elements, quantify the represented values and determine a relative ordering of these statistical values. We test our model on the recently released FigureQA dataset, which provides images and accompanying questions, for bar graphs and pie charts, augmented with rich annotations. Our approach outperforms the state-of-the-art Relation Networks baseline and traditional CNN-LSTM models when evaluated on this dataset. Our model also has a considerably faster training time of approximately 2 days on 1 GPU compared to the Relation Networks baseline which requires around two weeks to train on 4 GPUs.