EpiCon: Collective Agent Learning through Co-Evolving Multimodal Memory

2026-09-29

AI Summary

This study addresses the challenges of multi-agent experience reuse and cross-modal memory sharing bottlenecks by proposing a parameter-update-free shared multimodal memory framework. The framework introduces a novel architecture comprising two independently trained lightweight models—a 2B-parameter memory controller and a tree-based self-organizer—that jointly refine textual and visual evidence, enable collaborative experience evolution, and perform hierarchical integration through a hierarchical retrieval mechanism. Experimental results demonstrate that the proposed method achieves macro-average score improvements of up to 4.9 points across eleven benchmarks while reducing memory operation time by 74%, substantially enhancing both processing efficiency and performance for multimodal tasks.

Abstract

Agents can learn from past executions, but enabling different agents to reuse and build on one another's experience remains challenging. We introduce EpiCon, a shared multimodal memory framework for agent collective learning without updating host model parameters. EpiCon links question-level memory evolution to a persistent experience bank through two independently trained 2B models: a memory controller and a tree self-organizer. The controller jointly refines textual guidance and visual evidence across attempts and selectively includes visual memory. The self-organizer consolidates lessons hierarchically and retrieves experience and rules for new problems. We evaluate EpiCon on eleven benchmarks spanning four multimodal task domains, using two harnesses and multiple backbones. A frozen bank improves other systems even with a single solving attempt. A second harness raises the original system's macro-average score by 2.6 points across eleven benchmarks. Across four host configurations, EpiCon improves macro-average scores by 1.7 to 4.9 points over No Memory and reduces memory-operation time by 67\% to 74\% relative to backbone-sized memory models.

Problem

• Collective Agent Learning
• Multimodal Memory
• Experience Reuse
• Multi-Agent Collaboration

Innovation

• Collective Agent Learning
• Multimodal Memory
• Memory Controller
• Tree Self-Organizer
• Parameter-Free