Grant: Discovery Grants Program - Individual
University of Toronto
$27,232.00
Integrating Information Over Time in Human Brain Networks
FY 2016-17
Summary
OVERVIEW: We aim to understand how information is integrated over time within the human brain. Integrating information over time is the process by which prior information automatically modulates the meaning of later input. For example, it is how the earliest words in this sentence shape the meaning of words that arrive later on. It is also a process that occurs over many timescales: each word gains full meaning in the context of a sentence, and each sentence in the context of a paragraph. Given that the information we gather at each moment is most meaningful in relation to prior events, the neuroscientific question is: how are neural circuits organized so that prior information continuously shapes our understanding of the present?
Experiments using functional magnetic resonance imaging (fMRI) and electrocorticography (ECoG) have revealed three basic characteristics of how temporal integration is neurally implemented (Hasson et al, 2008; Lerner et al, 2011; Honey et al, 2012; Stephens et al, 2013). First, temporal integration is a distributed process: history-dependence is observed in the neural dynamics of almost all cortical regions. Second, temporal integration is hierarchically organized: the earliest levels of the sensory hierarchy respond on a moment-by-moment basis, while higher-level cortical regions respond based on information accumulated over increasingly long timescales (seconds or minutes). Third, regions of the cerebral cortex that integrate information over longer timescales also exhibit intrinsically slower neural dynamics.
We propose two lines of inquiry to develop our emerging model of neocortical temporal integration:
OBJECTIVE 1: Speech Perception and Production Over Time. Our existing model accounts only for temporal integration during perception, and not during behavior. To extend our model to describe the temporal organization of behavior, we will record neural activity as people perceive and reproduce long speech segments. The first question is: how is temporal integration performed in the production system? The second question is: which regions are shared between the perceptual and production systems, and how does information flow within and between the two hierarchies?
OBJECTIVE 2: Computational Modeling of Temporal Integration. We will distill our observations into a computational model that can be compared against our data from Objective 1. With this model, we address two questions. First, can hierarchically ordered temporal integration arise from the same learning mechanisms that give rise to hierarchical representations of spatial information? Second, how do higher-order regions maintain incoming information over time, enabling the behavioral reproduction of perceptual input?
IMPACT: The proposed work impacts natural sciences and engineering in three ways. First, we address the ability of neural circuits to integrate information over time; this is a fundamental phenomenon in mammalian neuroscience, which is poorly understood (Buonomano & Maass, 2009) and particularly important in light of the large signal variance associated with these processes (Leopold et al, 2003; Honey et al, 2012). Second, the modeling component of this proposal will bridge our knowledge of how temporal integration is performed in the human brain and how it is implemented in multi-level machine learning architectures. The representation of time-varying contextual information is a flagship problem in machine learning (Bengio, 2009). Finally, this work strengthens the Canadian skills base by training highly qualified personnel (graduate and undergraduate students) in experimental methodology, high-dimensional data analysis, machine learning, and scientific writing.
Program
Discovery Grants Program - Individual
Recipient
Location
Ontario, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
RGPIN-2014-04465
Name nom
Honey, Christopher
Department département
Arts and Science, Faculty of
Organization id
31
Institution établissement
University of Toronto
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2016
Competition year année de concours
2014
Award amount
27232
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
Discovery grants programs
Group fr
Programmes de subventions à la découverte
Committee code
1502
Committee name en
Biological Systems and Functions
Committee name fr
Systèmes et fonctions biologiques
Area of application code
1200
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Advancement of knowledge
Area of application fr
Avancement des connaissances
Research subject code
9999
Research subject group en
Not available
Research subject group fr
Non disponible
Research subject en
Not available
Research subject fr
Non disponible
Partie
3
Nb partie
5
Application title
Integrating Information Over Time in Human Brain Networks
Keyword
NSERC
Application summary
OVERVIEW: We aim to understand how information is integrated over time within the human brain. Integrating information over time is the process by which prior information automatically modulates the meaning of later input. For example, it is how the earliest words in this sentence shape the meaning of words that arrive later on. It is also a process that occurs over many timescales: each word gains full meaning in the context of a sentence, and each sentence in the context of a paragraph. Given that the information we gather at each moment is most meaningful in relation to prior events, the neuroscientific question is: how are neural circuits organized so that prior information continuously shapes our understanding of the present?
Experiments using functional magnetic resonance imaging (fMRI) and electrocorticography (ECoG) have revealed three basic characteristics of how temporal integration is neurally implemented (Hasson et al, 2008; Lerner et al, 2011; Honey et al, 2012; Stephens et al, 2013). First, temporal integration is a distributed process: history-dependence is observed in the neural dynamics of almost all cortical regions. Second, temporal integration is hierarchically organized: the earliest levels of the sensory hierarchy respond on a moment-by-moment basis, while higher-level cortical regions respond based on information accumulated over increasingly long timescales (seconds or minutes). Third, regions of the cerebral cortex that integrate information over longer timescales also exhibit intrinsically slower neural dynamics.
We propose two lines of inquiry to develop our emerging model of neocortical temporal integration:
OBJECTIVE 1: Speech Perception and Production Over Time. Our existing model accounts only for temporal integration during perception, and not during behavior. To extend our model to describe the temporal organization of behavior, we will record neural activity as people perceive and reproduce long speech segments. The first question is: how is temporal integration performed in the production system? The second question is: which regions are shared between the perceptual and production systems, and how does information flow within and between the two hierarchies?
OBJECTIVE 2: Computational Modeling of Temporal Integration. We will distill our observations into a computational model that can be compared against our data from Objective 1. With this model, we address two questions. First, can hierarchically ordered temporal integration arise from the same learning mechanisms that give rise to hierarchical representations of spatial information? Second, how do higher-order regions maintain incoming information over time, enabling the behavioral reproduction of perceptual input?
IMPACT: The proposed work impacts natural sciences and engineering in three ways. First, we address the ability of neural circuits to integrate information over time; this is a fundamental phenomenon in mammalian neuroscience, which is poorly understood (Buonomano & Maass, 2009) and particularly important in light of the large signal variance associated with these processes (Leopold et al, 2003; Honey et al, 2012). Second, the modeling component of this proposal will bridge our knowledge of how temporal integration is performed in the human brain and how it is implemented in multi-level machine learning architectures. The representation of time-varying contextual information is a flagship problem in machine learning (Bengio, 2009). Finally, this work strengthens the Canadian skills base by training highly qualified personnel (graduate and undergraduate students) in experimental methodology, high-dimensional data analysis, machine learning, and scientific writing.