Are you interested in pushing the boundaries of existing Large Language Models (LLMs)? Join our project ARAG, where you will be developing new Retrieval-Augmented Generation (RAG) pipelines that are Adaptive (ARAG) in the sense that they can decide on the actual architecture at runtime for a given query.
In this project, we will focus on addressing a critical challenge in LLMs: their inability to handle current and factual knowledge effectively. Many real-world queries to LLMs involve information that an LLMs may not have access to. A solution to this issue is RAG, and there are plenty of approaches to enhance LLM's reasoning capabilities one way or another. In this project group (PG), you will implement an RAG system that can choose a query-dependent architecture, i.e., for the given query, it can decide on the run which route to follow.
Your group will target three main components to handle ARAG: Retrieval (how additional information is actually retrieved), Augmentation (how retrieved information is passed on to an LLM), and Generation (how to generate the final response given a query and additional context from the previous steps).
The PG will be available in PAUL (TBA).
Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yi Dai, Jiawei Sun, Qianyu Guo, Meng Wang, Haofen Wang: Retrieval-Augmented Generation for Large Language Models: A Survey. CoRR abs/2312.10997 (2023) Wenqi Fan, Yujuan Ding, Liangbo Ning, Shijie Wang, Hengyun Li, Dawei Yin, Tat-Seng Chua, Qing Li: A Survey on RAG Meeting LLMs: Towards Retrieval-Augmented Large Language Models. KDD 2024: 6491-6501