Complete & Orthogonal Replication of Hyperdimensional Memory via Elementary Cellular Automata

Hyper-dimensional computing (HDC)/ Vector Symbolic Architectures (VSA) [1] implements associative learning using very large binary vectors. This approach has been used to model the learning and transfer learning of a foraging honeybee [2]. In real-world systems, we may want to copy learned associations onto multiple agents, e.g. swarm systems; however, simply copying the memory vectors across multiple agents makes all agents vulnerable to the same attack by a malicious entity. Therefore the challenge is to replicate the parent agent’s item memory and compositional memory such that all learned associations are preserved yet the clone’s memory vectors are maximally uncorrelated with the parent’s memory vectors. This work evaluated all 256 elementary cellular automata (ECA) rules for this task and identified 8 rules that satisfied these replication requirements. To the best of the authors’ knowledge, this is the first report of complete and orthogonal replication of HDC memory using ECA.