Modeling the main components of the neural dynamics during visual perceptual learning
Perceptual learning is an important form of adaption in sensory systems and beyond. Changes in the neural dynamics in response to repetitively solving a visual task has been shown to occur in the visual cortex of awake monkeys. However, how the modifications of the temporal dynamics of a network of many neurons can be understood remains vastly unclear. Here, we record simultaneously from a large number of neurons in the visual cortex of awake monkeys while performing a visual pattern detection task. The monkey has to find an elongated contour in one of two presented stimuli consisting of otherwise randomly oriented short bars. Neural spiking responses were recorded over a time of 2-3 weeks of learning to perform this task.
By using components analysis and dimensional reduction techniques derived from the machine learning community, we first show that the high dimensional neural population responses can be described by the dynamics of only two main components. (1) A facilitatory component involving cells having receptive fields nearby the contour to be detected, and (2) a suppressive component involving cells having receptive fields in the surround. We describe further the temporal response dynamics of these components and find that their strength depends on the length of the contour to be detected and that suppressive activity systematically lags the facilatory drive. Furthermore, we found that the weight of these components change during the course of learning. The weighting of both components is increased over the days and correlates well with performance of the task.
To better understand the mechanism of the learning induced network changes we devised a simple neuron network model. We found that a simple rate-based model can replicate the general dynamics of the two components thereby suggesting a neural mechanism for how perceptual learning might induced changes to the network activity.