Abstract: Motivated by problems in recommendation systemsand bioinformatics, we consider the problem of completing a lowrank, partially observed binary matrix with graph information.We show that the corresponding problem can be set up in apositive and unlabeled data learning (referred to asPU learninginliterature) framework. We make connections to convex optimiza-tion and show the existing greedy methods can be used to solve theproblem. Experiments on simulated data as well as gene-diseaseassociations data from bioinformatics show that using graphs,and adapting matrix completion in the PU learning setting, yieldadvantages over the standard binary matrix completion
- PU Matrix Completion with Graph Information
N. Natarajan, N. Rao, I. Dhillon.
International Workshop on Computational Advances in Multi-Sensor Adaptive Processing (CAMSAP), December 2015.