Inhibitory gating of non-linear dendrites enables stable learning of assemblies without forgetting

Neuronal assemblies — groups of neurons with strong recurrent connectivity — are thought to be the basic building blocks of perception and memory in the brain: representations of specific concepts. Activation of an assembly signals the presence of its associated concept to downstream targets. An efficient implementation should allow similar concepts to be represented by similar assemblies, sharing a subset of neurons. However, learning overlapping assemblies in computational models is difficult. Similar assemblies tend to either merge or diverge during the learning process. The same mechanism also leads to a problem known as `catastrophic forgetting', where the learning of a new assembly can disrupt the existing structure and thus lead to `forgetting' of previously learned assemblies. We propose a combination of biologically motivated mechanisms as a solution to these challenges: A neuron model with non-linear dendritic dynamics, inhibition enabling context-dependent dendritic gating, and spike-timing-dependent plasticity. These components allow for learning stable, overlapping neuronal assemblies for a variety of distinct tasks without catastrophic forgetting. Furthermore, by learning stable associations, our assembly framework can carry out diverse computations.