Abstract: We consider a setting when a study or source population for individualized-treatment-rule (ITR) learning can differ from the target population of interest. We assume subject covariates are available from both populations, but treatment and outcome data are only available from the source population. Existing methods use "importance" and/or "overlap" weights to adjust for the covariate differences between the two populations. We develop a general weighting framework that allow a better bias-variance trade-off than existing weights. Our method seeks covariate balance over a non-parametric function class characterized by a reproducing kernel Hilbert space. Our weights encompasse the importance weights and overlap weights as special cases. Numerical examples demonstrate that our weights can improve many ITR learning methods for the target population that rely on weighting.