Machine Learning

Cognitive Modeling focuses on decisions made within what Allen Newell described as the "cognitive band", short timeframes on the order of seconds and fractions of seconds, and current cognitive theories excel at modeling decisions at this grain size.  However, there are time scales above this and below this where modeling is often necessary to capture phenomena completely.  Further, decisions are often situated in some environment, and perceptual and motor interaction is necessary for these decisions to play out.  In these areas, we rely on techniques taken from the field of Machine Learning to fill in these important gaps.  Thus, decision tree induction, instance-based modeling, latent semantic analysis, and dimension reduction are all important tools in our modeling toolbox.  These tools all focus on how to learn from a dataset, where in our case that dataset is primarily a collection of behavioral traces we would like to model.

An added advantage of our fluency in Machine Learning techniques is the ability to model behavior in cases where behavioral fidelity doesn't matter, or when the underlying thought process isn't as important as just imitating the behavior (this is especially true of models that won't be used to make predictions in other domains or situations).  Sometimes it's more important to have a very rough approximation of behavior functioning approximately in some environment quickly, and  we apply the techniques mentioned above to these cases when they are called for.