AI seminar: Using novel articulated creatures to study perceptual learning of biological motion

Speaker: Daniel Saunders (Queens University)

There has been much debate about the nature of the processes involved in biological motion perception. In contrast to previous views, Jastorff et al. (2007) provided evidence that biological motion is understood via a mechanism that is specialized, but also relatively plastic, alterable over the course of hours rather than years. Rather than being limited to known animals, it is capable of efficiently learning novel motion patterns if they have certain properties in common with the locomotion of animals on the earth.

The challenge with studying biological motion in a learning paradigm is that most human and animal body shapes and how they move are already very well learned. In this presentation I will discuss my ideas for how virtual creatures ("aliens'') can be created using simulated evolutionary processes (through existing software packages) with body structures that are unlike any earth animals, but nevertheless possess a rigid skeleton and a locomotion pattern that is efficient relative to the laws of physics. This will allow me to dissociate, for the first time, the role of familiarity from the role of a skeletal architecture and obedience to physics in how easy a motion pattern is to perceive.

Bio: Daniel Saunders completed his undergraduate degree in Computer Science with the Cognitive Science option at the University of Waterloo in 2003. During his degree he worked as a research assistant for Professor Paul Thagard for a term, and coauthored a book chapter with him. Since that time he has pursued his Masters and now Ph.D. in Psychology at Queens University, primarily within the field of perception of biological motion. He also has interests in evolutionary psychology, coordination and communication, creativity, and cognitive modellling.