Telling time with recurrent networks in the brain

RODRIGO LAJE; DEAN V. BUONOMANO

Panamá

Encuentro; 2012 Pew Annual Meeting; 2012

The Pew Charitable Trusts

The brain?s ability to tell time and to produce precisely timed movements is critical to a wide range of tasks, like playing a musical instrument, dancing, or playing a video game. Yet, we remain mostly ignorant of the neural mechanisms underlying timing on the range of up to a few seconds?even at the simplest temporal task, like distinguishing the duration of two sounds. Generally speaking, the representation of temporal information in the brain remains one of the most elusive concepts of neurobiology. The dominant view has been that timing relies on centralized and specialized neural circuits dedicated to process time across different modalities. The most well known instantiation of this view is the internal clock model, where a dedicated group of neurons functions as a pacemaker, while another group implements an accumulator that counts the pulses emitted by the pacemaker. Evidence on the nature or the location of the pacemaker or the accumulator, however, has been elusive. An opposing view is that timing is such a critical part of brain function that most neural circuits are intrinsically capable of temporal processing?meaning that brain sensory, motor, association, or frontal areas can ?tell time? if the task at hand demands it. We propose that time is inherently encoded in the complex, continuously changing pattern of self-sustained activity of a recurrent neural network in any of these areas?a ?population clock? with no explicit measure of time. A number of predictions derive from this framework, which we implement as a computational model and test with psychophysical experiments. For instance, consider a subject?s production of a sequence of consecutive time intervals (a rhythm or temporal pattern) by finger tapping. Perhaps counterintuitively, the model successfully predicts that the temporal pattern as a whole is timed continuously from the beginning, instead of timing independently every interval in the sequence as in a ?reset? strategy.