Tsinghua Releases Comprehensive Survey on Memory Architectures in LLMs

THU-KEG · hf · 2026-07-30

THU-KEG published a systematic survey on memory mechanisms in large language models (LLMs). The paper introduces an architecture-centric taxonomy characterizing memory along three orthogonal axes: representation (implicit vs explicit), update dynamics (offline vs online), and persistence (short-term vs long-term). This framework bridges disparate architectural paradigms and analyzes system-level efficiency trade-offs.

Original post →

More from Research

Research channel →