Abstract
The recognition of complex movements and actions is a fundamental visual capability. In a series of psychophysical and functional imaging studies we have investigated the role of learning for the recognition of biological motion. Subjects were able to learn the discrimination between artificial novel biological motion stimuli. fMRI results indicate that several visual areas are involved in this learning process. More specifically, lower-level motion-related areas (hMT+/V5 and KO/V3B) show an emerging sensitivity for the differences between the discriminated stimuli, and higher-level areas (STS and FFA) show an increase of sensitivity after training. In addition, we find an overall reduction of the BOLD activity after training. Based on a hierarchical physiologically-inspired neural model for biological motion recognition (Giese & Poggio, 2003) we tried to reproduce the BOLD signal changes during discrimination learning. We show that learning of novel templates for complex movement patterns, which are encoded by sequences of body shapes and optic flow patterns, can be implemented by hebbian learning. We propose a model for the learning process that combines competitive and time-dependent hebbian plasticity, implementing physiologically plausible local learning rules. Our results demonstrate that these mechanisms can account for the emerging sensitivity for novel movement patterns observed in fMRI. We conclude that our model provides a well-defined framework to test different hypotheses about the mechanisms of neuronal plasticity underlying the learning of novel biological movements.
DFG, German Volkswagen Foundation, Hermann Lilly-Schilling Foundation