Preprint
Aug 2026
Dual-Grained Agent Memory and Shapley Context Attribution for Multimodal Agentic Learner
DG-Mem is proposed, a dual-grained agentic memory framework that augments a frozen MLLM with a non-parametric, externally stored memory built once from training-time rollouts and consulted read-only at test time, and improves consistently over no-memory and competitive memory baselines.
Jieke Wang, Tiancheng Shen, Yibo Yang et al.
· 0 citations