Dean Buonomano

Department of Neurobiology
Brain Research Institute
University of California, Los Angeles
(March 15, 2016)

The Neural Basis of Timing and Temporal Processing

As children, we learn to jump rope, carefully timing our next jump to the expected time it will take for the rope to make another revolution. But how does the brain keep track of temporal information, such as how long it might take the rope to make it back to our feet? Rather than a small clock residing in one particular brain area, Dr. Buonomano discussed the role of recurrent activity in networks of neurons as a method of “time-keeping.” His lab has demonstrated that even neurons in a slice of brain in a dish can learn to “tell” time based on a temporal pattern. The recurring activity within the network can lend to the anticipation of a future, expected event.

The brain’s ability to seamlessly assimilate and process temporal information is critical to most behaviors, from understanding speech to anticipating events such as when a traffic light will change colors. Because timing and temporal processing represent a fundamental neural computation, we postulate that there is no single brain area responsible for timing but rather that most neural circuits are intrinsically able to process temporal information on an as-needed basis. We have proposed that most forms of timing bear little resemblance with man-made clocks and, rather, rely on the neural dynamics of recurrent neural networks. Specifically, that sensory timing (e.g., interval discrimination) emerges from interaction of the time-varying internal state of neural networks with external stimuli. The internal state is defined not only by ongoing activity (the active state) but by time-varying synaptic properties, such as short-term synaptic plasticity (the hidden state). Psychophysical support for this hypothesis was presented.

In contrast to sensory timing, motor timing requires networks to actively generate responses, and thus rely on self-perpetuating activity within neural networks. It has been proposed that motor timing emerges from the dynamics of balanced recurrent neural networks. But a long-standing limitation of this theory is that the relevant regime is chaotic. We show that it is possible to tame chaos in firing rate recurrent networks, by tuning the recurrent weights. Within this framework, networks store spatiotemporal objects (such as handwritten words) as the voyage through phase space — as opposed to the destination (i.e., a fixed point attractor). One prediction of the theory that timing is an intrinsic property of neural networks is that even in vitro cortical slices may be able to “learn” to tell time. We provide support for this prediction by demonstrating that intervals chronically presented to cortical circuits through optogenetic stimulation can “learn” simple temporal patterns.