Multimodal Agent MemorySubmit a paper

PersonaVLM

PersonaVLM: Long-Term Personalized Multimodal LLMs

Chang Nie, Chaoyou Fu, Yifan Zhang, Haihua Yang, Caifeng Shan · 2026-03-20

Read the paperPDF

Management
T2 Dynamic Memory
Memory changes during deployment through a stable management policy.
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. chronological multimodal memory summaries

    1. Represented asPerception-to-text records & summariesCross-represented explicit
    2. Stored inHierarchical multi-tier store
    3. Retrieved bySimilarity search, Metadata / spatiotemporal retrieval
    4. Source evidenceSource linked & recoverable
    The system proactively extracts and summarizes chronological multimodal interactions.Full paper: Update · PDF p. 2 · Figure 1

  2. personalized memory database

    1. Represented asSpatial, temporal & cross-modal structured memoryCross-represented explicit
    2. Stored inHierarchical multi-tier store
    3. Retrieved byPlanner / tool-assisted retrieval
    4. Source evidenceSource linked & recoverable
    Consolidated memories and evolving personality information are organized for multi-turn retrieval and response alignment.Full paper: 3.1. PersonaVLM Framework · PDF p. 4 · Figure 1

How memory changes (T2 operations)

  • Update · Prompted LLM/VLM record update

    A fixed extraction-and-consolidation process updates the personalized database after interactions. For PersonaVLM, this specifically supports LLM-mediated memory editing of chronological multimodal memory summaries.

  • Consolidation · Episodic / hierarchical summarization

    Chronological multimodal interactions are summarized before storage. For PersonaVLM, this specifically supports episodic or hierarchical summarization of chronological multimodal memory summaries.

  • Consolidation · Experience abstraction

    Long-term interaction evidence is abstracted into durable user characteristics. For PersonaVLM, this specifically supports abstraction of episodes into reusable experience of chronological multimodal memory summaries.

Related systems