Multimodal Agent MemorySubmit a paper

Evo-MedAgent

Evo-MedAgent: Beyond One-Shot Diagnosis with Agents That Remember, Reflect, and Improve

Weixiang Shen, Bailiang Jian, Jun Li, Che Liu, Johannes Moll, Xiaobin Hu, et al. · 2026-04-15

Read the paperPDF

Management
T3 Self-Evolving Memory
Feedback during deployment changes the memory-management policy itself.
Primary modality
Image
Also: text
Representation
Cross-represented explicit

How experience becomes memory

Each path is one memory component, traced from what is stored to how it is found again.

  1. retrospective clinical episodes

    1. Represented asPerception-to-text records & summariesCross-represented explicit
    2. Stored inVector-indexed collection
    3. Retrieved bySimilarity search
    4. Source evidenceSource linked & recoverable
    Past multimodal cases are stored as retrospective episodes for similar-case retrieval.Full paper: Memory Architecture · PDF pp. 3–4 · Figure 1

  2. procedural heuristics and tool reliability

    1. Represented asSpatial, temporal & cross-modal structured memoryCross-represented explicit
    2. Stored inModular adapter bank
    3. Retrieved byPlanner / tool-assisted retrieval
    4. Source evidenceSource not retained
    Priority-tagged diagnostic rules and per-tool trust scores guide future diagnostic reasoning.Full paper: Memory Architecture · PDF pp. 3–4 · Figure 1

How memory changes (T2 operations)

  • Consolidation · Experience abstraction

    Reflection converts completed cases into reusable diagnostic heuristics. For Evo-MedAgent, this specifically supports abstraction of episodes into reusable experience of retrospective clinical episodes.

  • Revision · Evidence-weighted belief revision

    Observed tool outcomes update per-tool trustworthiness. For Evo-MedAgent, this specifically supports evidence-weighted revision of stored belief or confidence of retrospective clinical episodes.

How the policy evolves (T3)

  • Retrieval policy

    Test-time outcomes persistently recalibrate tool trust and heuristic priority, changing which reusable evidence and procedures are selected in later cases. The persistent later-use effect is deployment-feedback evolution of the reusable retrieval policy or procedure.

Related systems