Formation of Artificial Neural Assemblies by Biologically Plausible Inhibition Mechanisms

A research team led by Lucas Hoff and Gustavo Soroka has published a significant paper titled 'Formation of Artificial Neural Assemblies by Biologically Plausible Inhibition Mechanisms' on arXiv. The work addresses a key limitation in existing Assembly Calculus (AC) models, which use a rigid k-winners-take-all selection process that doesn't reflect the statistical power-law scaling of real neural activity. To create more biologically realistic artificial intelligence, the team developed a new model called E%-WTA.

This model introduces two major innovations: an E%-winners-take-all selection mechanism inspired by the dynamics of gamma oscillation cycles in the brain, and an inhibition process based on the actual ratio of excitatory to inhibitory neurons observed across the cerebral cortex. These changes allow the network's own dynamics—rather than a preset parameter—to determine the size of the neural assemblies that form. The results show these biologically-inspired assemblies have a superior recovery rate when evoking the stimuli that created them, outperforming the original AC model. This represents a meaningful step toward AI systems whose internal organization and learning mechanisms more closely resemble those of biological brains.