Synaptic low-rank modulation facilitates adaptation in cortical networks

The human brain quickly adapts to a rapidly changing environment, often faster than what is thought to be possible by way of synaptic plasticity. Real-time control adaptation can only be achieved through faster processes such as neuromodulation. A number of neuromodulation models based on gain regulation have been proposed recently. But most of them utilize neuron-wide regulation and can not accommodate the simultaneous impact of different types of neurotransmitters on different parts of the dendritic tree. Models of synaptic gating and connectivity-tuning demonstrate that using enough additional parameters, one can mold a desired task into the network. Nevertheless, the relationship between the minimal amount of modulation and dimensionality of the task variation is not understood. Here, we address both of these questions using tractable representations of synaptic neuromodulation in recurrent neural networks (RNNs). Consistent with the experimental findings, modulation is performed by an external network of modulatory neurons, adapting their strengths and direction of modulation in response to the task variation. We introduce fixed synaptic masks, representing susceptibilities of the individual synapses to the neurotransmitter release from each modulatory neuron. These masks form a scaffold for neuromodulation patterns that can be applied to the network through activation of corresponding modulatory neurons. We observe that appropriate combination of masks can effectively mold network connectivity, adapting it to the changed environment. Specifically, we demonstrate how synaptic modulation can facilitate performance of the network on varying tasks of pattern discrimination and detection and link the (minimum) number of modulatory controls to the degrees of freedom in task variation. Our approach creates a useful framework for improving the mechanistic understanding of rapid adaptation in neural circuits. We shift the focus from synaptic plasticity as learning mechanism for one task to flexible neuromodulation that fits an entire subspace of tasks.