Neurons as Monte Carlo Samplers: Sequential Bayesian Inference in Spiking Neural Populations
Animals constantly face the problem of estimating unknown world states from noisy and ambiguous sensory input in a dynamical environment. Neuro-psychophysical experiments have suggested brains handle these sensory uncertainties using approximate Bayesian inference. This indicates that neurons both maintain the probability distribution over world states and combine prior knowledge of the environment with likelihood of sensory observations in a probabilistic manner. Here we explore how spiking neurons can implement a form of approximate Bayesian inference known as particle filtering. The time-varying posterior probability distribution of hidden world states is represented directly by spikes in a population of neurons, without involving complicated decoding methods. Each spike represents a sample of a particular world state. Each possible world state is represented by an ensemble of nearby identically tuned neurons.Spikes across the entire population approximate the complete posterior probability distribution. Neural variability in spiking arises naturally as a consequence of sampling during inference. The posterior spike distribution is recursively updated by an array of neurons, each of which integrates feed-forward Poisson spike trains whose intensities represent the likelihood of sensory measurements, with recurrent inputs representing propagation of previous posterior probability distribution, according to Bayes' rule. Since Bayesian calculation requires neurons to multiply probabilities of previous posterior and new sensory measurements, we show how such multiplication can be carried out approximately by a variety of neurons, ranging from simple LIF neurons to detailed four-compartment pyramidal neurons. Model parameters were selected to be consistent with those reported for typical CNS neurons. We present simulations that demonstrate how spikes in a neural population might encode the dynamic environmental state whose values follow a continuous trajectory over time. The framework we introduce here can also be extended to Bayesian inference over other types of probabilistic graphical models.
Conference:
Computational and systems
neuroscience 2009, Salt Lake City, UT, United States, 26 Feb - 3 Mar, 2009.
Presentation Type:
Poster Presentation
Topic:
Poster Presentations
Citation:
(2009). Neurons as Monte Carlo Samplers: Sequential Bayesian Inference in Spiking Neural Populations.
Front. Syst. Neurosci.
Conference Abstract:
Computational and systems
neuroscience 2009.
doi: 10.3389/conf.neuro.06.2009.03.048
Copyright:
The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers.
They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.
The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.
Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.
For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.
Received:
30 Jan 2009;
Published Online:
30 Jan 2009.