ABSTRACT

This chapter introduces the use of local, non-parametric methods for dimension reduction of human performance data. It focuses on trajectory-fitting and explores a best-fit trajectory is the most basic kind of reduced-dimension action model. The chapter discusses how the work, and how their use of local information helps build good trajectory models from human performance data. Trajectory fitting generates a one-dimensional action model, with a parameterization representing the temporal ordering of points in the best-fit trajectory. The problem with the approach of growing a higher-dimensional data representation from the “skeleton” of a best-fit trajectory is the issue of whether the meaning of the second or third feature score is similar in different regions of the model. The local direction of the curve at its approximation of a given example point will likely be very different than the direction of the trajectory in configuration space from which the point was sampled.