I think there's some interest in learning optimal decision trees in the community, as well as robust learning methods under different kinds of adversarial influence. They're less open problems and more areas of potential improvement though.