Synaptic and nonsynaptic plasticity approximating probabilistic inference

Abstract

Learning and memory operations in neural circuits are believed to involve molecular cascades of synaptic and nonsynaptic changes that lead to a diverse repertoire of dynamical phenomena at higher levels of processing. Hebbian and homeostatic plasticity, neuromodulation, and intrinsic excitability all conspire to form and maintain memories. But it is still unclear how these seemingly redundant mechanisms could jointly orchestrate learning in a more unified system. To this end, a Hebbian learning rule for spiking neurons inspired by Bayesian statistics is proposed. In this model, synaptic weights and intrinsic currents are adapted on-line upon arrival of single spikes, which initiate a cascade of temporally interacting memory traces that locally estimate probabilities associated with relative neuronal activation levels. Trace dynamics enable synaptic learning to readily demonstrate a spike-timing dependence, stably return to a set-point over long time scales, and remain competitive despite this stability. Beyond unsupervised learning, linking the traces with an external plasticity-modulating signal enables spike-based reinforcement learning. At the postsynaptic neuron, the traces are represented by an activity-dependent ion channel that is shown to regulate the input received by a postsynaptic cell and generate intrinsic graded persistent firing levels. We show how spike-based Hebbian-Bayesian learning can be performed in a simulated inference task using integrate-and-fire (IAF) neurons that are Poisson-firing and background-driven, similar to the preferred regime of cortical neurons. Our results support the view that neurons can represent information in the form of probability distributions, and that probabilistic inference could be a functional by-product of coupled synaptic and nonsynaptic mechanisms operating over several timescales. The model provides a biophysical realization of Bayesian computation by reconciling several observed neural phenomena whose functional effects are only partially understood in concert.

A Spiking Working Memory Model Based on Hebbian Short-Term Potentiation

“…We use an AdEx IAF neuron model with spike-frequency adaptation (Brette and Gerstner, 2005) that was modified recently (Tully et al, 2014) for compatibility with a custom-made BCPNN synapse model in NEST through the addition of the intrinsic excitability current I β j (see Spike-based Bayesian learning rule). The model was simplified by excluding the subthreshold adaptation dynamics.…”

“…Plastic AMPA and NMDA synapses are modeled with a spike-based version of the Bayesian Confidence Propagation Neural Network (BCPNN) learning rule (Wahlgren and Lansner, 2001; Tully et al, 2014, 2016). For introductory purposes, we only highlight a few key equations here.…”

“…For introductory purposes, we only highlight a few key equations here. For a full derivation of the learning rule from Bayes rule, deeper biological motivation, and proof of concept, see Tully et al (2014). The E trace, which is critical for allowing delayed reward learning, has been omitted because such conditions are not applicable here.…”

DNA methylation and the regulation of stimulus-specific memory formation

“…local translation), homeostatic plasticity acts globally on the whole cell or neuronal network. Indeed, neuronal network models suggest that homoeostatic plasticity is important for preventing accelerating activity (Tully et al, 2014;Zenke et al, 2015) and allows the network to remain plastic.…”

“…Hebbian plasticity favours increased excitation and stably increased synaptic connections, but on the other hand the brain needs to remain a flexible system with a set overall activity. Recent models of neuronal networks highlight the importance of homoeostatic plasticity in keeping this balance (Tully et al, 2014;Zenke et al, 2015). Homeostatic plasticity describes changes in response to altered activity levels (e.g.…”