Neural dynamics in adaptive sensorimotor behavior
At any given instant, hundreds of billions of cells in our brains are lighting up in a complicated yet highly coordinated manner to give rise to our thoughts, percepts, and movements. A single neuron may be connected to thousands of other cells, sending out and receiving information through electrical impulses called spikes. From an engineering perspective, these spikes form a signal that may be viewed as a series of ones and zeros rapidly unfolding in time. Altogether, these signals reflect the ongoing computations taking place inside the nervous system, and as such, constitute a window into the brain’s inner workings. The overarching aspiration of my research is to understand how information that controls and optimizes our behavior is embedded in these complex patterns of neural activity, and how those patterns change during the process of learning.
A central function of the brain is the capacity to anticipate future sensorimotor events based on past experience. Picture, for instance, a batter preparing to strike an incoming baseball thrown at lightning speed. A naive player’s attempt at intercepting the ball is almost guaranteed to fail; there is simply not enough time to initiate the movement while processing the looming object. After a few failed trials, however, the player becomes increasingly close to striking the ball, and ultimately hits a home run. This begs the question: what has changed? At some level, information about the past throws must have been stored in the brain to allow the individual to hone in on the correct timing and finally hit the ball. But how does this happen with only a few lines of ones and zeros exchanged between neurons? And how do those complicated patterns of activity need to change when suddenly the opponent throws the ball at a different speed? This is ultimately the sorts of questions my thesis research aims to provide an answer to.
Because baseball is not really an option in a laboratory setting, I opted for a computerized version of the game. I train monkeys to intercept balls sliding across a computer screen and thrown at various speeds. Understandably, animals initially perform imperfectly at the task. Overtime, however, just like the inexperienced batter eventually learns to master the timing of the movement, monkeys become experts and may even surpass human performance. They do so by learning the range of pre-selected speeds that the ball can be thrown at. Ultimately, the animals become so attuned to the speed range, they can anticipate the trajectory of the ball with high precision. By analyzing patterns of activity shared across several hundreds of neurons while the animals are performing this task, I investigate how the dynamics of neural responses can adaptively store information about the speed range and progressively optimize the animals' behavior. Some of the results were recently published in the journal Neuron and featured in MIT news.
At any given instant, hundreds of billions of cells in our brains are lighting up in a complicated yet highly coordinated manner to give rise to our thoughts, percepts, and movements. A single neuron may be connected to thousands of other cells, sending out and receiving information through electrical impulses called spikes. From an engineering perspective, these spikes form a signal that may be viewed as a series of ones and zeros rapidly unfolding in time. Altogether, these signals reflect the ongoing computations taking place inside the nervous system, and as such, constitute a window into the brain’s inner workings. The overarching aspiration of my research is to understand how information that controls and optimizes our behavior is embedded in these complex patterns of neural activity, and how those patterns change during the process of learning.
A central function of the brain is the capacity to anticipate future sensorimotor events based on past experience. Picture, for instance, a batter preparing to strike an incoming baseball thrown at lightning speed. A naive player’s attempt at intercepting the ball is almost guaranteed to fail; there is simply not enough time to initiate the movement while processing the looming object. After a few failed trials, however, the player becomes increasingly close to striking the ball, and ultimately hits a home run. This begs the question: what has changed? At some level, information about the past throws must have been stored in the brain to allow the individual to hone in on the correct timing and finally hit the ball. But how does this happen with only a few lines of ones and zeros exchanged between neurons? And how do those complicated patterns of activity need to change when suddenly the opponent throws the ball at a different speed? This is ultimately the sorts of questions my thesis research aims to provide an answer to.
Because baseball is not really an option in a laboratory setting, I opted for a computerized version of the game. I train monkeys to intercept balls sliding across a computer screen and thrown at various speeds. Understandably, animals initially perform imperfectly at the task. Overtime, however, just like the inexperienced batter eventually learns to master the timing of the movement, monkeys become experts and may even surpass human performance. They do so by learning the range of pre-selected speeds that the ball can be thrown at. Ultimately, the animals become so attuned to the speed range, they can anticipate the trajectory of the ball with high precision. By analyzing patterns of activity shared across several hundreds of neurons while the animals are performing this task, I investigate how the dynamics of neural responses can adaptively store information about the speed range and progressively optimize the animals' behavior. Some of the results were recently published in the journal Neuron and featured in MIT news.