Artificial IntelliDance: Teaching Machine Learning through a Choreography

In this paper we present a choreography that explains the process of supervised machine learning. We present how a perceptron (in its dual form) uses convolution kernels to learn to differentiate between two categories of objects. Convolution kernels such as string kernels and tree kernels are widely used in Natural Language Processing (NLP) applications. However, the baggage associated with learning the theory behind convolution kernels, which extends beyond graduate linear algebra, makes the adoption of this technology intrinsically difficult. The main challenge in creating this choreography was that we were required to represent these mathematical equations at their meaning level before we could translate them into the language of movement. By orchestrating such a choreography, we believe, we have obviated the need for people to posses advanced math background in order to appreciate the core ideas of using convolution kernels in a supervised learning setting.