IAG: Induction-Augmented Generation Framework for Answering Reasoning Questions

The article discusses a new framework called Induction-Augmented Generation (IAG) for answering reasoning questions. This framework utilizes inductive knowledge and retrieved documents for implicit reasoning. The authors propose two versions of IAG, namely IAG-GPT and IAG- Student. IAG-GPT utilizes the knowledge generated by GPT-3 for answer prediction, while IAG- Student incorporates a student inductor model. The article concludes that IAG outperforms Retrieval-Augmented Generation (RAG) baselines and ChatGPT in Open-Domain QA tasks.

Publication date: 1 Dec 2023
Project Page: https://www.huawei.com
Paper: https://arxiv.org/pdf/2311.18397