Semi-supervised learning for a ranking problem. (A) The query is a single point in input space, and the remaining points comprise the database one wishes to rank. (B) The ranking induced by Euclidean distance. Marking sizes are proportional to the ranking of each point. (C) The ideal ranking. Clearly, to find the optimal ranking we need to find the cluster/manifold structure in the data.
Weston et al. BMC Bioinformatics 2006 7(Suppl 1):S10 doi:10.1186/1471-2105-7-S1-S10