memn2n-keras

Memory Networks for Question Answering based on "End to End Memory Networks" Sukhbaatar et al - https://arxiv.org/pdf/1503.08895.pdf

This project was implemented and submitted as part of the project for the NLP course in UMass (Fall 2016). This project tries to train a single Neural Network model to solve all the 20 QA bAbI tasks in the Facebook's bAbI dataset (https://research.fb.com/downloads/babi/). We were able to reproduce the QA part of the paper with the results quite close to the paper's results. We were not able to experiment with the multi-hop models to solve the complex tasks, and the Language Modelling tasks due to restrictions in time and computing resources.

Languages used:

1. Python

Libraries used:

1. Keras - Deep Learning

2. Numpy - Matrix operations