A Hierarchical Semantic Memory model for classification of Motion Capture data

This paper describes a novel approach for supervised classification of motion capture data. A Hierarchical Semantic Memory (HSM) transforms a skeleton wireframe input sequence and learns a hierarchical representation based on space time similarities. HSM has several advantages as a classifier, such as minimal preprocessing of data, single-scan learning of training set, high classification accuracy with just one instance per class in the training set, and a short running time.