EngramNCA: a Neural Cellular Automaton Model of Memory Transfer

2025-04-16

AI Summary

This work addresses the limitation of synaptic plasticity in constraining memory storage and transmission within artificial systems. To overcome this, we propose a decentralized morphological memory model grounded in intracellular implicit memory channels. Methodologically, we design EngramNCA—a novel engram-based neural cellular automaton—featuring, for the first time, a dual-channel state architecture comprising explicit visible states and private “genetic” memory channels. We further introduce synergistic GeneCA and GenePropCA modules to enable end-to-end differentiable morphological training via implicit memory modulation. Experiments demonstrate that the model stably generates, transmits across generations, and hierarchically coexists with complex morphological structures—without centralized control. On MNIST-derived seeds, it spontaneously emergently produces diverse morphologies, thereby validating a novel paradigm of non-synaptic memory encoding, transfer, and multi-morphological self-organization.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Machine Learning: Deep Generative Models & Autoencoders
• Multiagent Systems: Modeling other Agents

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

This study introduces EngramNCA, a neural cellular automaton (NCA) that integrates both publicly visible states and private, cell-internal memory channels, drawing inspiration from emerging biological evidence suggesting that memory storage extends beyond synaptic modifications to include intracellular mechanisms. The proposed model comprises two components: GeneCA, an NCA trained to develop distinct morphologies from seed cells containing immutable"gene"encodings, and GenePropCA, an auxiliary NCA that modulates the private"genetic"memory of cells without altering their visible states. This architecture enables the encoding and propagation of complex morphologies through the interaction of visible and private channels, facilitating the growth of diverse structures from a shared"genetic"substrate. EngramNCA supports the emergence of hierarchical and coexisting morphologies, offering insights into decentralized memory storage and transfer in artificial systems. These findings have potential implications for the development of adaptive, self-organizing systems and may contribute to the broader understanding of memory mechanisms in both biological and synthetic contexts.

Problem

• Modeling memory transfer using neural cellular automata
• Integrating visible and private memory channels in NCAs
• Exploring decentralized memory storage in artificial systems

Innovation

• Neural cellular automaton with private memory channels
• GeneCA and GenePropCA for morphology control
• Decentralized memory storage and transfer mechanism