Beyond the Context Window: A Survey of Retrieval-Augmented Memory for Large Language Model Enhancement

PhD Qualifying Examination


Title: "Beyond the Context Window: A Survey of Retrieval-Augmented Memory for 
Large Language Model Enhancement"

by

Mr. Zhengjun HUANG


Abstract:

Retrieval-Augmented Generation (RAG) has become a central paradigm for
extending large language models beyond the intrinsic limitations of fixed
context windows and static parametric knowledge. By retrieving external
evidence at inference time, RAG reframes LLMs as reasoning and synthesis
engines that can consult an explicit, updateable knowledge source, thereby
improving factual grounding and domain adaptability. As research and
applications have matured, retrieval augmentation has progressed from a
predominantly flat, vector-based pipeline to increasingly structured and
experience-aware memory systems, giving rise to two prominent subsequent
paradigms: GraphRAG and agentic memory. This survey reviews this trajectory
from classical RAG to GraphRAG and beyond. We review the evolution from RAG
to GraphRAG and beyond, covering dense and interaction-based retrieval,
corpus-structured graph retrieval for global reasoning, and persistent
episodic memory for cross-session continuity. This survey unifies these
paradigms and highlights emerging directions such as multimodal memory and
retrieval-augmented intelligence.


Date:                   Wednesday, 5 February 2026

Time:                   4:30pm - 6:30pm

Venue:                  Room 2132C
                        Lift 22

Committee Members:      Prof. Xiaofang Zhou (Supervisor)
                        Prof. Raymond Wong (Chairperson)
                        Dr. Yangqiu Song