Synapses learn to utilize pre-synaptic noise for the prediction of postsynaptic dynamics

Abstract

Synapses in the brain are highly noisy, which leads to a large trial-by-trial variability. Given how costly synapses are in terms of energy consumption these high levels of noise are surprising. Here we propose that synapses use their noise to represent uncertainties about the activity of the post-synaptic neuron. To show this we utilize the free-energy principle (FEP), a well-established theoretical framework to describe the ability of organisms to self-organize and survive in uncertain environments. This principle provides insights on multiple scales, from high-level behavioral functions such as attention or foraging, to the dynamics of single microcircuits in the brain, suggesting that the FEP can be used to describe all levels of brain function. The synapse-centric account of the FEP that is pursued here, suggests that synapses form an internal model of the somatic membrane dynamics, being updated by a synaptic learning rule that resembles experimentally well-established LTP/LTD mechanisms. This approach entails that a synapse utilizes noisy processes like stochastic synaptic release to also encode its uncertainty about the state of the somatic potential. Although each synapse strives for predicting the somatic dynamics of its neuron, we show that the emergent dynamics of many synapses in a neuronal network resolve different learning problems such as pattern classification or closed-loop control in a dynamic environment. Hereby, synapses coordinate their noise processes to represent and utilize uncertainties on the network level in behaviorally ambiguous situations.