Efficient computational algorithms are what made graphical Markov models so popular and successful. Similar algorithms can also be developed for computation with compositional models, which form an alternative to graphical Markov models. In this paper we present a theoretical basis as well as a scheme of an algorithm enabling computation of marginals for multidimensional distributions represented in the form of compositional models.
Physical exercise instruction sheets are difficult to understand. In general, considerable information is hidden in these types of instruction sheets, which also makes them difficult for machines to understand. Major missing information types include the source and destination location of a human movement. Here we present a Bayesian network to extract the implicit or missing information from typical exercise instruction sheets. We proposed two different kind of Bayesian networks which consists of three and four variables respectively. The network with three variable are designed to for single exercise instruction with single action or pose and the other one designed for single or multiple sentence with two actions or poses. The conditional probability table (CPT) is the backbone of the Bayesian network. At the start, the CPT is updated from our physical exercise instruction sheet corpus (PEISC). Keeping the Action and Bodypart fixed, we have developed our CPT using a unique approach, i.e., crowdsourcing, where we have developed a CPT update system using 13 different exercises consisting of 44 different exercise videos. Using this system based on the rating of a participant of the video the specific variable of that CPT is updated automatically in the Bayesian network. We also updated the Action variable, which consists of 14 different values (action verbs) using crowdsourcing with a human computation approach.