My curriculum vitae (2/2/12).
Some of my projects--in pictures (made in R). Note that these are preliminary rather than final results. Nonetheless, if something piques your interest, e-mail me!
A project with Richard Veale (that we spawned in a course taught by Olaf Sporns) investigating the effects of network structure and input topology on the computational power of biologically-inspired spiking neural networks (i.e., reservoirs/liquid state machines).
Poster here, presented at the 2011 Midwest Cognitive Science Conference in Michigan.
Using a game paradigm allows us to record moment-to-moment decisions during a trial--including changes of mind. Thus, we can do both traditional correctness and response time analyses, as well as trajectory analyses to better distinguish decision-making models and multiple process accounts. By varying the speed of the falling objects, we can also adjust the deadline to respond. Also, games are fun: our participants often thank us for making a fun experiment--and want to know if their score was good!
A project with Jennifer Trueblood and John Kruschke investigating how decision-makers (DMs) combine advice from expert advisers when the DMs know how much unique evidence (i.e., medical tests) each adviser has access to, but not the test results each adviser saw. If advisers agree that some outcome is likely, but are using different information, then a rational DM will extremify, combining the independent evidence to make a stronger rating. If one adviser sees a superset of the test results that another adviser sees, a rational DM should just match the rating of the adviser with greater knowledge. If advisers have unique knowledge and give different advice, a rational DM will compromise. We offer a Bayesian information aggregation model that produces these normative behaviors, and find that it works really well--at least for good learners in our task! A journal publication is forthcoming.
Expanding on the question of how word-object pair frequency affects learning, I ran a few experiments in which I gave learners four blocks of the same cross-situational training and testing. These learning trajectories for pairs of different frequency span roughly 30 minutes, and provide models with nuanced details to describe. Large individual differences--evident even after the first block of training--are a source of variance we hope to explain using the trajectory data. A publication is forthcoming.
Work with Caitlin Fausey, Drew Hendrickson, and Rob Goldstone about learning and transfer effects of verbal labels. A model is being born!
Extending my current model with a new working-memory mechanism.
Can't say much yet, but this is exciting!
Do learners first look to new objects or old ones? What does a look tell us about what they know? Multilevel logistic regression to the rescue!