Multimodal Agent MemorySubmit a paper

PolarMem

PolarMem: A Training-Free Polarized Latent Graph Memory for Verifiable Vision-Language Models

Zhisheng Chen, Tingyu Wu, Zijie Zhou, Zhengwei Xie, Jinhan Li, Ziyan Weng, et al. · 2026-05-30

Read the paperPDF

Management
T1 Static Memory
Memory is built once and reused without change during deployment.
Primary modality
Image
Also: text
Representation
Hybrid
Symbolic positive and negative constraints are attached to visual nodes that can be returned as evidence.

How experience becomes memory

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

  1. visual nodes and local evidence tokens

    1. Represented asRetained perceptual source evidenceSource- and modality-specific
    2. Stored inGraph / relational store, Vector-indexed collection
    3. Retrieved bySimilarity search, Structure traversal
    4. Source evidenceSource directly available
    Visual nodes retain holistic visual semantics and local visual evidence; retrieved visual nodes are assembled as visual evidence tokens for the frozen VLM.Full paper: Polarized Latent Graph Memory · PDF p. 5

  2. HAS, NOT_HAS, and uncertain concept states

    1. Represented asSpatial, temporal & cross-modal structured memoryCross-represented explicit
    2. Stored inGraph / relational store
    3. Retrieved byStructure traversal, Similarity search
    4. Source evidenceSource linked & recoverable
    The heterogeneous graph stores verified presence and explicit absence as separate edge types; lexicographic retrieval filters logical conflicts before semantic ranking.Full paper: Polarized Latent Graph Memory · PDF p. 5 · Algorithm 1

Related systems