Exploiting Feature Heterogeneity for Improved Generalization in Federated Multi-task Learning | IEEE Conference Publication | IEEE Xplore