Interests
I am currently a postdoctoral researcher in the Psychology Department at Rutgers University, working with Jacob Feldman in the Visual Cognition Lab. My interests include human concept learning, feature selection, and inductive inference.
Projects in Progress
- Complexity-dependence of feature selection in human concept learning
- There is a general assumption in much of the concept-learning and causal-learning literature that feature selection (i.e., an observer's choice of features or attributes on which to base classification or causal inference) is guided by feature informativeness. This assumption however neglects any role for the complexity of the relationships by which features are related to one another or to class variables. We are therefore exploring the role of relational complexity in feature selection and utilization.
- Incremental Gaussian mixture-model simulation of one-dimensional concept learning
- The most popular models of human concept learning are at present the so-called exemplar models or instance learners, such as GCM, which perform a summed-similarity computation between a new observation and prior examples to determine the classification of the new datum. While such models have been successful in accounting for concept learning performance in humans under various conditions, their heavy memory load, lack of internal conceptual model, and intrinsic ability to learn concepts of arbitrarily high complexity (a capacity not shared by humans) would appear to limit their attractiveness as models of human learning. We are investigating an incremental exemplar/mixture-model hybrid learner for one-dimensional and two-dimensional feature spaces to address these issues.
Publications and Reports
- Human Sensitivity to Mutual Information
- Fass, David, Ph.D. Thesis, Rutgers University, 2006.
Categorization Under Complexity: A Unified MDL Account of Human Learning of Regular and Irregular Categories
Fass, D. and Feldman, J., Advances in Neural Information Processing Systems, Proceedings of the 2002 Conference, edited by S. Becker, S. Thrun, and K. Obermayer, Vol. 15, MIT Press, Cambridge, MA, 2003.
On the Role of Category Complexity in Concept Learning
Fass, David, Masters Thesis, Rutgers University, 2004.
Approximation of Discrete Multivariate Probability Distributions: Recursive and Hierarchical Approaches
A survey of some interesting literature.
