Cecile G. Tamura
The brain can’t directly encode the passage of time, but recent work hints at a workaround for putting timestamps on memories of events.
https://bit.ly/2N6PDc0
The brain can’t directly encode the passage of time, but recent work hints at a workaround for putting timestamps on memories of events.
https://bit.ly/2N6PDc0
"As sensory neurons fire in response to an unfolding event, the brain
maps the temporal component of that activity to some intermediate
representation of the experience — a Laplace transform, in mathematical
terms.
That representation allows the brain to preserve information about the event as a function of some variable it can encode rather than as a function of time (which it can’t). The brain can then map the intermediate representation back into other activity for a temporal experience — an inverse Laplace transform — to reconstruct a compressed record of what happened when." https://bit.ly/1vfaaPE
Other scientists independently uncovered neurons, dubbed “time cells,” that were “as close as we can possibly get to having that explicit record of the past. These cells were each tuned to certain points in a span of time, with some firing, say, one second after a stimulus and others after five seconds, essentially bridging time gaps between experiences. Scientists could look at the cells’ activity and determine when a stimulus had been presented, based on which cells had fired. This was the inverse-Laplace-transform part of the researchers’ framework, the approximation of the function of past time. “
“A second can last forever. Days can vanish. It’s this coding by parsing episodes that, to me, makes a very neat explanation for the way we see time. We’re processing things that happen in sequences, and what happens in those sequences can determine the subjective estimate for how much time passes.”
That timeline could be of use not just to episodic memory in the hippocampus, but to working memory in the prefrontal cortex and conditioning responses in the striatum.
Scientists also started to show that the same equations that the brain could use to represent time could also be applied to space, numerosity (our sense of numbers) and decision-making based on collected evidence — really, to any variable that can be put into the language of these equations. “For me, what’s appealing is that you’ve sort of built a neural currency for thinking, If you can write out the state of the brain … what tens of millions of neurons are doing … as equations and transformations of equations, that’s thinking."
One day cognitive models could even lead to a new kind of artificial intelligence built on a different mathematical foundation than that of today’s deep learning methods. Only last month, scientists built a novel neural network model of time perception, which was based solely on measuring and reacting to changes in a visual scene.
But before any application to AI is possible, scientists need to ascertain how the brain itself is achieving this.
That representation allows the brain to preserve information about the event as a function of some variable it can encode rather than as a function of time (which it can’t). The brain can then map the intermediate representation back into other activity for a temporal experience — an inverse Laplace transform — to reconstruct a compressed record of what happened when." https://bit.ly/1vfaaPE
Other scientists independently uncovered neurons, dubbed “time cells,” that were “as close as we can possibly get to having that explicit record of the past. These cells were each tuned to certain points in a span of time, with some firing, say, one second after a stimulus and others after five seconds, essentially bridging time gaps between experiences. Scientists could look at the cells’ activity and determine when a stimulus had been presented, based on which cells had fired. This was the inverse-Laplace-transform part of the researchers’ framework, the approximation of the function of past time. “
“A second can last forever. Days can vanish. It’s this coding by parsing episodes that, to me, makes a very neat explanation for the way we see time. We’re processing things that happen in sequences, and what happens in those sequences can determine the subjective estimate for how much time passes.”
That timeline could be of use not just to episodic memory in the hippocampus, but to working memory in the prefrontal cortex and conditioning responses in the striatum.
Scientists also started to show that the same equations that the brain could use to represent time could also be applied to space, numerosity (our sense of numbers) and decision-making based on collected evidence — really, to any variable that can be put into the language of these equations. “For me, what’s appealing is that you’ve sort of built a neural currency for thinking, If you can write out the state of the brain … what tens of millions of neurons are doing … as equations and transformations of equations, that’s thinking."
One day cognitive models could even lead to a new kind of artificial intelligence built on a different mathematical foundation than that of today’s deep learning methods. Only last month, scientists built a novel neural network model of time perception, which was based solely on measuring and reacting to changes in a visual scene.
But before any application to AI is possible, scientists need to ascertain how the brain itself is achieving this.
No comments:
Post a Comment