Abstract:Maximum mean discrepancy (MMD) has been widely adopted in domain adaptationto measure the discrepancy between the source and target domain distributions.Many existing domain adaptation approaches are based on the joint MMD, which iscomputed as the (weighted) sum of the marginal distribution discrepancy and theconditional distribution discrepancy; however, a more natural metric may betheir joint probability distribution discrepancy. Additionally, most metricsonly aim to increase the transferability between domains, but ignores thediscriminability between different classes, which may result in insufficientclassification performance. To address these issues, discriminative jointprobability MMD (DJP-MMD) is proposed in this paper to replace thefrequently-used joint MMD in domain adaptation. It has two desirableproperties: 1) it provides a new theoretical basis for computing thedistribution discrepancy, which is simpler and more accurate; 2) it increasesthe transferability and discriminability simultaneously. We validate itsperformance by embedding it into a joint probability domain adaptationframework. Experiments on six image classification datasets demonstrated thatthe proposed DJP-MMD can outperform traditional MMDs.

Discriminative Joint Probability Maximum Mean Discrepancy (DJP-MMD) for Domain Adaptation