[초청강연] From Neurons to Perception

From Neurons to Perception

Much of sensory neuroscience has focused on how the brain faithfully represents sensory information. However, perception is not a faithful representation of sensory inputs: Rather, it is an inference, i.e., an interpretation of sensory evidence based on the brain’s prior expectations about the sensory world.
To infer a perceptual object, sensory evidence belonging to a common perceptual object must be bound together. At the neuronal level, spikes signaling a common perceptual object must be bound in time by firing synchronously. In the first part of my talk, I will show that the temporal context for “binding” is provided by a distinct subtype of inhibitory interneurons, coined “grnsFS” (Shin & Moore, 2019?Neuron). These grnsFS neurons spike regularly at ~40Hz rhythmicity, also known as gamma rhythmicity. Increased regularity and increased gamma spiking of grnsFS predicted successful detection of subtle sensory stimuli, suggesting their role in successful perception. Much like a CPU clock that provides time slots for computation in a computer, grnsFS provides time slots for synchronous spiking amongst excitatory neurons.
In the second part of my talk, I will show how illusions can be utilized to study perception as an inference. Illusions provide a unique context where the inferred percept noticeably and reproducibly deviates from the physical reality. Traditionally, perceptual inference, e.g., illusory percepts, is thought to emerge in higher visual areas. In contrast, primary visual cortex (V1) is thought to contain a faithful representation of visual inputs. Using an artificial neural network to decode 1000’s of simultaneously recorded neurons, I show that V1 neural activity is sufficient to infer an illusory contour. Most of this “inference” signal is contained in a small subset (~14%) of neurons that respond to illusory contours as they would to real edges of the same orientation. Ongoing efforts are directed at elucidating the necessary and sufficient mechanisms for illusory contour encoding in V1, both at the microcircuit level (e.g., inhibitory mechanisms in V1) and at the mesoscale level (e.g., higher visual areas feedback to V1). Future studies will also test the hypothesis that grnsFS is necessary for illusory contour perception.

Hyeyoung Shin
Postdoctoral Scholar, Adesnik Lab, University of California, Berkeley
Ph.D. in Neuroscience, Moore & Jones Labs, Brown University