The functional role of oscillatory dynamics in neocortical circuits: A computational perspective by Felix Effenbergera, Pedro Carvalhoa, Igor Dubinina, Wolf Singera ISBN 10.1073/PNAS.2412830122 instant download

The functional role of oscillatory dynamics in neocortical circuits: A computational perspective by Felix Effenbergera, Pedro Carvalhoa, Igor Dubinina, Wolf Singera ISBN 10.1073/PNAS.2412830122 instant download

The dynamics of neuronal systems are characterized by hallmark features suchas oscillations and synchrony. However, it has remained unclear whether thesecharacteristics are epiphenomena or are exploited for computation. Due to thechallenge of selectively interfering with oscillatory network dynamics in neuronalsystems, we simulated recurrent networks of damped harmonic oscillators in whichoscillatory activity is enforced in each node, a choice well supported by experimentalfindings. When trained on standard pattern recognition tasks, these harmonic oscillatorrecurrent networks (HORNs) outperformed nonoscillatory architectures with respectto learning speed, noise tolerance, and parameter efficiency. HORNs also reproduceda many characteristic features of neuronal systems, such as the cerebral cortexand the hippocampus. In trained HORNs, stimulus-induced interference patternsholistically represent the result of comparing sensory evidence with priors stored inrecurrent connection weights, and learning-induced weight changes are compatiblewith Hebbian principles. Implementing additional features characteristic of naturalnetworks, such as heterogeneous oscillation frequencies, inhomogeneous conductiondelays, and network modularity, further enhanced HORN performance withoutrequiring additional parameters. Taken together, our model allows us to give plausiblea posteriori explanations for features of natural networks whose computational role hasremained elusive. We conclude that neuronal systems are likely to exploit the uniquedynamics of recurrent oscillator networks whose computational superiority criticallydepends on the oscillatory patterning of their nodal dynamics. Implementing theproposed computational principles in analog hardware is expected to enable the designof highly energy-efficient and self-adapting devices that could ideally complementexisting digital technologies.