Trade-off between reconstruction loss and feature alignment for domain generalization | IEEE Conference Publication | IEEE Xplore