Mutual information scheduling for ranking

In this paper, an efficient and flexible scheduling algorithm for estimating the ranking of n objects through pairwise comparisons is proposed. By successively picking pairs for comparison such that the mutual information between the parameter(s) of interest and the outcome of the comparison is maximized, we quickly obtain a good approximation of these parameters using few comparisons. In simulation, this scheduling algorithm performs better than randomly selecting pairs for comparison or selecting pairs according to a natural heuristic previously proposed. However, a main advantage to this principled approach is that it appropriately adapts to a variety of models and parameters.