EpiCon: Collective Agent Learning through Co-Evolving Multimodal Memory

arXiv:2609.37923 · cs.CV · Submitted 2026-09-29 · Read on arXiv

cs.CV

Submitted: 2026-09-29

Updated: 2026-09-29

Comments: Preprint

Code: https://github.com/deepseek-ai/deepseek-harness

Project page: https://zzzmyyzeng.github.io/EpiCon

License: http://creativecommons.org/licenses/by/4.0/

The gist: Agents can learn from past executions, but enabling different agents to reuse and build on one another's experience remains challenging.

Terminology

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.

Sources

Related papers