|
The goal of this project is to integrate two powerful research methodologies to provide further insights as to the nature of human cognition. Brain imaging, research using the recent advances o functional magnetic resonance imaging (fMRI) allows us to observe which brain regions are active in the performance of cognitive tass. Cognitive modeling allows us to predict the behavior of subjects in complex intellectual tasks. The brain imaging work has been strong in identifying brain regions that serve specific functions, but weak in understanding integrated cognitive function. The cognitive modeling has allowed us to understand integrated and complex function but has been weak in identifying what brain regions are involved and in providing identifiably between alternative models. This project will bring together researchers who represent distinct and successful cognitive modeling architectures (ACT-R: Anderson, Soar: Lewis, PDP: McClelland, SAC: Reder, CAP2: Schneider) with researchers who are at the forefront of fMRI imaging (Carter, Fiez, Schneider) with the goal of tightly relating cognitive function in the models to human cortical dynamics. A set of complex tasks will be identified that involve various aspects of learning, problem solving, and language, and for which models exist in at least one of the architectures represented by our group. Brain imaging studies will be conducted with the same set of subjects performing all of these tasks. The complex tasks will include: rule learning, strategy learning, fan effect, tower of Hanoi, mathematical problem solving, sentence processing (embedding, and ambiguity), and congruity effects. To provide further constraints, the same set of subjects will also be imaged while performing some simpler and more traditional 'pure' tasks, including: N-back, attention switching, word processing, declarative encoding, executive monitoring/error detection and checkerboard phase lag retinotopy mapping. Models will be developed in the various cognitive architectures which address both the brain imaging data and behavioral data from these tasks. The brain imaging methodology will involve both blocked designs, in which aggregate patterns of activation are compared across conditions, and event related designs (ERN), in which the activation functions are synchronized across time to specific events in the experiment. Complex tasks are particularly appropriate for ERN designs in fMRI because they have a long time course appropriate to the temporal resolution of fMRI. Our approach to relating brain activation and computational models of cognition will involve seven steps: a) create a model that predicts the behavioral data; b) have the simulation create an event log of critical cognitive operations; c) utilize hemodynamic delay equations to predict the family of activation structures; d) characterize the fMRI activation functions based on the types of activations seen in the model (e.g., present during learning but not during practiced responding, stimulus locked, response locked); 3) see which brain activation patterns map the model and which do not; f) revise the model to account for the activation data; and g) try the same model for a new task to evaluate whether the predicted model structures can predict activation patterns with no new free model parameters except for the task specification (e.g., productions). This research will contribute to the brain-imaging literature by moving us beyond localization and towards an understanding of how distinct areas participate in large circuits to perform the integrative computations required for complex cognition. The research contributes to the cognitive modeling literature by providing a new anchor to compare architectures and by helping to resolve the identifiably problems that have haunted these architectures when there was only behavioral data for comparison. The research will also contribute to cognitive science generally by merging multiple constraints in the understanding of cognition and by eliminating methodological barriers among researchers. A major effect will be made to educate researchers from both fields about the results of this integrative effort. The behavioral and brain imaging data, model simulations, computerized experimental procedures, as well as the modeling and analysis tools will be made available on the web. This functional human activation/modeling data set will provide an empirical test bed for comprehensive modeling of human global brain function. As a final potential contribution, an understanding o the neural basis of complex cognition has the potential of making brain research relevant to issues of education in the critical areas of skill acquisition and learning.
Key Personnel
|