Description
- Abstract:
- Humans are remarkably adept at generalizing knowledge between experiences in a way that can be difficult for computers. Often, this entails generalizing constituent pieces of experiences that do not fully overlap, but nonetheless share useful similarities with previously acquired knowledge. A young musician may learn to play several instruments in different contexts, for example, learning the flute to play classical and the saxophone to play jazz. Because the fingerings of the saxophone and the flute are nearly the same, a musician can re-use previously learned hand motions for different effect across the two instruments. Conversely, a musician may learn to play a single song across many instruments that require completely distinct physical motions, but yet still transfer knowledge between them. This degree of compositionality is critical for flexible goal-directed behavior but can be difficult for computational frameworks because they often assume an underlying structure of the world that is incompatible with generalization. Here, I propose a novel computational framework that leverages the compositional structure in real-world environments by assuming people generalize goals, or what to do, independently from a word model, or how to do it. I examine the computational framework normatively and ask when it makes sense to generalize goals and world models independently or together. In a series of experiments, I compare human subject behavior to model predictions, showing that people adapt their generalization strategy depending on the environment. Together, these results sug- gest that no one strategy is best across all environments, and that while it is adaptive to represent a complex environment in learnable components, people pay attention to the relationship between components in their environment.
- Notes:
- Thesis (Ph. D.)--Brown University, 2018
Access Conditions
- Rights
- In Copyright
- Restrictions on Use
- Collection is open for research.
Citation
Franklin, Nicholas Thompson,
"Compositionality in Human Structure Learning"
(2018).
Cognitive, Linguistic, and Psychological Sciences Theses and Dissertations.
Brown Digital Repository. Brown University Library.
https://doi.org/10.26300/9bkf-a791
Relations
Collection:
-
Cognitive, Linguistic, and Psychological Sciences Theses and Dissertations
Theses and Dissertations for the Cognitive, Linguistic, and Psychological Sciences department....