Nicholas Barendregt, Department of Applied Mathematics, University of Colorado Boulder

Adaptive Decision-Making in Dynamic Environments Using Sequential Bayesian Inference

In natural environments, individuals must develop robust strategies to make accurate and timely decisions to ensure their survival. In simple environments, a normative description of these strategies can be derived analytically by assuming that individuals act as Bayes-optimal observers. However, complexities in environmental dynamics, like unpredictable changes in task conditions, can lead to behavior that deviates from the predictions of classical theory. As a result, researchers have turned to heuristic, sub-optimal models to understand experimental data. In these more naturalistic settings, normative decision models and their relevance to behavior and cognition are not well understood. In this dissertation, I will extend normative theory to consider strategies for accumulating evidence and making decisions in complex environments that exhibit temporal fluctuations in structure over the span of a single deliberative process. The resulting models adapt their evidence integration and decision commitment processes to the stochasticity of the environment, display rich dynamics, and can be efficiently utilized to generate testable predictions about behavior. Furthermore, by re-analyzing experimental data, I will show that adaptive normative models can better explain human behavior than commonly-used heuristics, demonstrating the importance of a normative perspective in understanding cognition.

Nicholas Barendregt, Department of Applied Mathematics, University of Colorado Boulder

Adaptive Decision-Making in Dynamic Environments Using Sequential Bayesian Inference

In natural environments, individuals must develop robust strategies to make accurate and timely decisions to ensure their survival. In simple environments, a normative description of these strategies can be derived analytically by assuming that individuals act as Bayes-optimal observers. However, complexities in environmental dynamics, like unpredictable changes in task conditions, can lead to behavior that deviates from the predictions of classical theory. As a result, researchers have turned to heuristic, sub-optimal models to understand experimental data. In these more naturalistic settings, normative decision models and their relevance to behavior and cognition are not well understood. In this dissertation, I will extend normative theory to consider strategies for accumulating evidence and making decisions in complex environments that exhibit temporal fluctuations in structure over the span of a single deliberative process. The resulting models adapt their evidence integration and decision commitment processes to the stochasticity of the environment, display rich dynamics, and can be efficiently utilized to generate testable predictions about behavior. Furthermore, by re-analyzing experimental data, I will show that adaptive normative models can better explain human behavior than commonly-used heuristics, demonstrating the importance of a normative perspective in understanding cognition.