EngramNCA: a Neural Cellular Automaton Model of Memory Transfer

TL;DR

EngramNCA introduces a dual-channel memory scheme for Neural Cellular Automata, pairing a public channel with a private intracellular memory to emulate biological memory engrams. The architecture comprises GeneCA, which learns primitive embeddings in private gene channels while updating public states, and GenePropCA, which modulates those private memories to generate richer morphologies; together as an ensemble, they enable hierarchical growth and memory transfer. Across experiments, the model achieves coexisting primitives, morphologies built from primitives, and multiple morphologies from a shared substrate, with Lenia-like moving patterns faithfully reproduced and stabilized beyond initial frames. These findings suggest that incorporating private memory channels into artificial self-organizing systems can enhance stability, modularity, and transfer of learned dynamics, with potential applications in adaptive AI and insights for neuroscience on intracellular memory substrates. The work also offers a computational framing for memory engrams and multi-timescale information transfer, guiding future explorations into more robust, self-organizing architectures and memory-inspired transfer mechanisms.

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.

Figures (26)

• Figure 1: NCA augmentation through gene embedding channels, where the channels of the cell are represented, including the RGB channels, the alpha channel, three additional channels, plus three gene channels (G1,G2, and G3) representing the private tape.
• Figure 2: Binary embedding of geometric morphologies and their respective encoding.
• Figure 3: Binary embedding of lizard body sections and their respective encodings.
• Figure 4: One step of the GeneCA
• Figure 5: One step of the GenePropCA