How do you make money on the stock market? Buy low, sell high! How do you known when the stock is low or high? You need to determine when the stock is changing direction. These direction changes are refered to as changepoints. Obviously there's a great deal of interest in detecting changepoints in time series data. Our focus is on applying and generalizing these techniques to domains in cognitive science.
Where does changepoint detection come into play in cognitive science? People have an ongoing stream of experience, and people constantly have to make the determination as to whether their previous experience applies to the current situation or not. Detecting changepoints -- points in time where there is a break from past experience and future experience -- is critical to this end. For example, suppose you have a spouse who is very attentive and caring and then starts to behave differently. Are they just having a bad day, or has something changed in their attitude toward you?
We are currently exploiting cool Bayesian models to attempt to infer not only changepoints and make predictions over time, but also to infer properties of the environment (e.g., how often changepoints occur).