Judging a person as a friend or foe, a mushroom as edible or poisonous, or a sound as an l or r are examples of categorization problems. One key aspect of learning is discerning the relevant stimulus dimensions that determine category membership and the value and costs (in terms of time, cognitive efort, and dollars) associated with gathering such information. Many category learning models employ selective attention mechanisms that learn which stimulus dimensions are most critical to performance. However, these models make the unrealistic assumption that all stimulus dimensions will be encoded, and, thus, fail to address challenges that arise from limited processing resources, both cognitive and neural. Improved models are required to understand the interplay between attentional allocation and memory. By recasting category learning as a dynamic decision process, we develop a model that selectively encodes information during learning as a function of the learner's goals, task demands, and knowledge state. To capture the required interplay between attention, memory, and executive function, our model consists of two primary components: one that determines the value of potential sources of information based on the decision maker's goals and assumptions about the world and a second component that reflects the decision maker's current knowledge. Current knowledge represented by the second learning component is utilized by the first value component to direct information gathering. The learning component of the model is updated by the information selected by the first component, completing the cycle of mutual influence. A central goal of the proposal is to develop models that make realistic assumptions about human capacity limitations and to characterize how individuals'mental machinery and behavioral outcomes deviate from rational principles. A second goal is to combine our novel model-based approach with eye tracking and functional magnetic resonance imaging (fMRI) to determine the neural mechanisms that support goal-directed attention and learning. Model-based analyses of fMRI data have the power to go beyond conventional analysis methods to reveal complex dynamics between neural systems that give rise to cognitive competencies. In two proposed studies, participants must decide which information sources to sample, taking into account the conflicting needs of (1) minimizing information cost, (2) making the correct decision, and (3) learning more about the categories and information sources with the aim of increasing performance on future trials. By fitting our model to individuals'information seeking and classification behavior, we can calculate a number of regressors that track unobservable mental states that are predictive of subsequent behavior and critical for determining the brain basis of the dynamic decision making processes that support category learning. Advancing our knowledge of the brain processes that underlie these powerful aspects of cognition may have real-world consequences by providing knowledge about optimal learning strategies as well as providing insight into disorders that affect learning and memory.
Impairments in learning, memory, and attention deficits accompany a number of psychiatric (e.g., schizophrenia, major depression, ADHD) and neurological disorders (e.g., Alzheimer's disease, epilepsy). Accordingly, understanding the neural mechanisms of attention and memory in the healthy brain promises to advance neurobiological theory and may lead to new developments that bear on the diagnosis and treatment of such conditions.
