Survey from 30+ institutions maps the evolution of multimodal agent memory on two axes
Fudan-University · hf · 2026-10-15
A large-scale survey led by Fudan University with 30+ institutions (including Meta, NVIDIA, and researchers like Metaxas and Choi) charts how multimodal agent memory should evolve for long-horizon, lifelong settings.
- Axis I (Formation & Representation): what information is retained and how it is encoded—modality-specific embeddings, textual/structured records, latent and parametric representations.
- Axis II (Memory Management): three tiers—T1 Static (fixed reuse), T2 Dynamic (stable update policy), T3 Self-Evolving (policy evolves with experience).
- The paper synthesizes recurring design patterns and principles: balancing source fidelity vs. abstraction, maintaining temporal validity and provenance, grounding memory use in recoverable evidence, and evolving management policies, plus open challenges.
More from Research
- Alibaba's HICO-Bench shows 36% of highly-rated agent conversations fail verification — Tongyi-ConvAI · 2026-10-15
- PixText fixes text in generated images: 83% glyph pass rate vs 37% for baselines — ATH-MaaS · 2026-10-15
- TextVerifier catches flawed Chinese glyph rendering with F1 0.844, far above OCR and VLMs — ATH-MaaS · 2026-10-15
- ByteDance's Diff Diffusion predicts code edit paths: 2.78x faster inference, 39% lower latency — Xiaolong Xu · 2026-10-15
- MimicX refines humanoid tracking with policy-in-the-loop feedback: 25.7% error cut — Shuaijun Liu · 2026-10-15
- MagicBeanStalk: 88% of RLVR data trapped in zero advantage, recursive synthesis adds 8.6 points — Lianhuiq · 2026-10-15