Summary of Re-rag: Improving Open-domain Qa Performance and Interpretability with Relevance Estimator in Retrieval-augmented Generation, by Kiseung Kim et al.
RE-RAG: Improving Open-Domain QA Performance and Interpretability with Relevance Estimator in Retrieval-Augmented Generation
by Kiseung Kim, Jay-Yoon Lee
First submitted to arxiv on: 9 Jun 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The Retrieval Augmented Generation (RAG) framework combines parametric and external knowledge to achieve state-of-the-art performance on open-domain question answering tasks. However, it suffers from degradation when queries are accompanied by irrelevant contexts. To address this, the RE-RAG framework introduces a relevance estimator that not only measures relative relevance but also provides confidence to classify whether given context is useful for answering the question. The RE is trained using weakly supervised methods and question-answer data without labels for correct contexts. Results show that RE improves small generator (sLM) fine-tuning and previously unreferenced large language models (LLMs). New decoding strategies utilize the proposed confidence, allowing users to know when an answer is “unanswerable” or relying on LLM’s parametric knowledge. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The RAG framework uses a combination of knowledge to do well on question-answering tasks. However, it struggles when there are irrelevant answers. To solve this problem, the RE-RAG framework adds a new way to measure relevance and how confident we should be in our answer. This is done by training a model using question-answer data without any extra information. The results show that this new approach works well for smaller models and even bigger ones. We can also use this confidence to make better choices, like telling the user when an answer isn’t possible or relying on what we already know. |
Keywords
» Artificial intelligence » Fine tuning » Question answering » Rag » Retrieval augmented generation » Supervised
