Refining Inductive Bias in Unsupervised Learning via Constraints

We propose the use of constraints as a technique for refining the inductive bias of unsupervised learning algorithms. Our previous work with a clustering algorithm and instance-level hard constraints has demonstrated that the use of constraints can both increase accuracy and decrease runtime. We here outline the other kinds of algorithms and constraints we intend to investigate. In addition, we present our intended method for evaluating constraint-based techniques.