1. Arun Baalaaji
2. Akash Gujju
3. Mukkesh Ganesh
4. Trisha Mandal
5. Balaji Chidambaram

One of the fundamental downstream task in NLP is Question Answering(QA). The QA problem provides a measurable and objective way to test the reasoning capacity of any intelligent systems. With the advancements in machine reasoning, a good natural language system should have the ability to perform multi-hop reasoning, where the system has to reason with information taken from more than one document to arrive at the answer. Most real-world questions require multi-hop reasoning to arrive at an answer. For multi-hop reasoning, a machine must understand the question, identify supporting facts from multiple knowledge sources and use reasoning to generate an answer. In this project, we want to focus on exploring various fusion techniques and experimenting with knowledge-based information retrieval systems.

We worked on top of the baselines found in public domain namely

1. DFGN(https://github.com/woshiyyya/DFGN-pytorch)
2. Hotpot model(https://github.com/hotpotqa/hotpot)

We made code changes on top of the forked code from the above repositories. For simpler use, we have cloned the reference codes with our edits on two directories baseline_code and dfgn_baseline. Refer utils for fine-tuning and fusion code.