Knowledge representation and retrieval

In this talk, I focus on one facet of my research program: knowledge retrieval. I formulate, describe, and extend a novel model that retrieves items by randomly following associations between items in memory. I show how this model can capture patterns in how people retrieve items from a category, patterns that previously were used to argue that memory search must be guided by a strategic, rather than a random process. Further, I show that for a random search over knowledge to capture human memory retrieval, knowledge must be represented in a structured manner (e.g., network), and that a spatial representation is insufficient. Extending the new model, I develop and empirically validate a novel machine learning method for estimating network representations of groups and individuals efficiently. I then apply this method to reveal differences between the knowledge representations of older individuals that are cognitively impaired and matched controls. I will conclude with a discussion of an in progress project, which examines healthy and unhealthy cognitive aging using a low-cost, naturalistic microlongitudinal design.