Domain adaptation on graphs by learning graph topologies: theoretical analysis and an algorithm

Abstract

Traditional machine learning algorithms assume that the training and test data have the same distribution,while this assumption does not necessarily hold in real applications. Domain adaptation methods take into account thedeviations in data distribution. In this work, we study the problem of domain adaptation on graphs. We consider a sourcegraph and a target graph constructed with samples drawn from data manifolds. We study the problem of estimating theunknown class labels on the target graph using the label information on the source graph and the similarity between thetwo graphs. We particularly focus on a setting where the target label function is learned such that its spectrum is similarto that of the source label function. We first propose a theoretical analysis of domain adaptation on graphs and presentperformance bounds that characterize the target classification error in terms of the properties of the graphs and the datamanifolds. We show that the classification performance improves as the topologies of the graphs get more balanced, i.e.as the numbers of neighbors of different graph nodes become more proportionate, and weak edges with small weightsare avoided. Our results also suggest that graph edges between too distant data samples should be avoided for goodgeneralization performance. We then propose a graph domain adaptation algorithm inspired by our theoretical findings,which estimates the label functions while learning the source and target graph topologies at the same time. The jointgraph learning and label estimation problem is formulated through an objective function relying on our performancebounds, which is minimized with an alternating optimization scheme. Experiments on synthetic and real data setssuggest that the proposed method outperforms baseline approaches.