|
François Leroy Research engineer (INSERM) Cognitive Developmental team Centre Neurospin, Unicog, CEA Université Paris-Saclay, France [email protected] |
|
I am currently studying brain predictions in infants with Shruti Naik and Ghislaine Dehaene-Lambertz. Neuroimaging tools: EEG, frequency tagging, attention modulations, decoding, Event-Related Variability (ERV) framework |
Research framework
Development of internal world models for updating expectations in a changing environment
“A primary function of the imagination would be to anticipate what is likely to happen
in light of stored information about past regularities (Harris, Child Development, 2021).”
Suppose a child is looking for her Christmas hat and can’t find it in the house. Suddenly, she hears the sound of the front door closing. She rushes to the next window and sees her brother leaving the house with her hat on. She may have solved the mystery by assembling several pieces of information into an internal model that incorporates the following elements: her hat is likely not in the house, her brother likes it and may have left the house.
An internal world model gathers prior knowledge about possible states of the world. In particular, it enables us to anticipate possible changes in the current environment (1). Such a computational model holds structured sets of priors that may include perceptual components with their behaviors or rules, i.e., together with a set of available probabilistic inferences (2).
I am interested in the many cognitive and flexible ways in which the brain combines prior knowledge and recently acquired sensory information into an internal world model to guide new expectations. When a child faces some change in his environment, his brain can anticipate the forthcoming environment, for example by generalizing auditory sequences (3), combining visual symbols (4) or spatially transforming line drawings (5).
An internal world model gathers prior knowledge about possible states of the world. In particular, it enables us to anticipate possible changes in the current environment (1). Such a computational model holds structured sets of priors that may include perceptual components with their behaviors or rules, i.e., together with a set of available probabilistic inferences (2).
I am interested in the many cognitive and flexible ways in which the brain combines prior knowledge and recently acquired sensory information into an internal world model to guide new expectations. When a child faces some change in his environment, his brain can anticipate the forthcoming environment, for example by generalizing auditory sequences (3), combining visual symbols (4) or spatially transforming line drawings (5).
Examples of generative models. A) Global rule in auditory sequences (3); B) Symbolic composition (4). “Look! Where is padu* duck?” * = two;
C) spatial transformation of line drawings: scene on the left and objects on the right (5)
Specifically, building an internal model of the world probably means
In addition, experimental paradigms must take into account the child’s motivation. Generative models often engage the child’s attention and require some executive control. Accurate expectations depend on confidence in one’s prior knowledge and the maturation of top-down sensory pathways. I will study “sweet spots” of predictability where information gain and curiosity balance the cost of such adaptive behavior (7).
Lastly, my research project is concerned with potential applications to the field of learning. My approach is in line with active learning by promoting the construction of internal models through local priors adjustment, incremental and systematic testing (8). Importantly, I will investigate to what extent these processes can be applied to multiple contexts of events. For example, the way in which the brain combines elements from several contexts can shed light on the transfer mechanisms of learning.
- Calling up resources through perception-specific top-down pathways despite the absence of sensory input.
- Simultaneously recruiting concepts, structures and inferences in a way similar to instantiating a schema (6);
- Transforming them into operational components using symbols, variable bindings, object files, image spaces, cognitive maps, etc.
- Finally, combining them flexibly into an expectation relevant to the future environment. Conscious access likely occurs during this stage.
In addition, experimental paradigms must take into account the child’s motivation. Generative models often engage the child’s attention and require some executive control. Accurate expectations depend on confidence in one’s prior knowledge and the maturation of top-down sensory pathways. I will study “sweet spots” of predictability where information gain and curiosity balance the cost of such adaptive behavior (7).
Lastly, my research project is concerned with potential applications to the field of learning. My approach is in line with active learning by promoting the construction of internal models through local priors adjustment, incremental and systematic testing (8). Importantly, I will investigate to what extent these processes can be applied to multiple contexts of events. For example, the way in which the brain combines elements from several contexts can shed light on the transfer mechanisms of learning.
- Diester, Ilka et al., Internal world models in humans, animals, and AI. Neuron, Volume 112, Issue 14, 2265 – 2268, 2024
- Tenenbaum, Joshua B. et al. How to Grow a Mind: Statistics, Structure, and Abstraction. Science, 331, 1279-1285, 2011
- Basirat Anahita et al. A hierarchy of cortical responses to sequence violations in three-month-old infants, Cognition, Volume 132, Issue 2, 137-150, 2014
- Pomiechowska, B. et al. Early-emerging combinatorial thought: Human infants flexibly combine kind and quantity concepts, Proc. Natl. Acad. Sci. U.S.A. 121 (29) e2315149121, 2024
- Dillon, Moira et al. Core geometry in perspective. Developmental Science. Vol 18, Issue 6, 2015
- Gilboa, Asaf et al. Neurobiology of Schemas and Schema-Mediated Memory. Trends in Cognitive Sciences, Volume 21, Issue 8, 618 - 631, 2017
- Kidd, C. et al. The Goldilocks Effect: Human Infants Allocate Attention to Visual Sequences That Are Neither Too Simple Nor Too Complex. PLOS ONE 7(5): e36399, 2012
- Bramley, Neil R. et al. Active inductive inference in children and adults: A constructivist perspective. Cognition, Volume 238, 2023
