The paper, "Distilling the Knowledge in a Neural Network" (Hinton et al. 2015) develops a methodology to transfer knowledge between classification learning models. Although knowledge transfer between models can be desirable in different practical scenarios, the paper highlights as one of the objectives achieving simpler models that obtain similar performance as cumbersome models or ensamble of specialist models without their computational overhead.
We are interested in modeling team strength over the length of a season. The strength metric will not be static as it evolves over the course of the season due to factors such as injuries, confidence and the ability of opposition teams to figure out how to play against a team. Therefore we are interested in modeling team strength as a time series.
The English Premier League contains 20 teams, which play each other home and away. The team strengths of these teams are modeled over the course of a season using state space methods. Using team strengths predictions for score differences are made for each game.