Fig. 13

(left) Movements recorded from stair ambulation, in red, form a distinct cloud mostly separable from those recorded from flat ground walking, in blue. (right) Movements recorded from 8 individuals occupy distinct annular regions in the latent representation. In³⁶ we show that a simple classifier can achieve 97% accuracy for distinguishing flat from stair walking, and 91% accuracy for discriminating amongst 8 individuals. These examples show that dimensionality reduction can assist in visualization, interpretation, or creating representations suitable for classification.