Risk-averse trees for learning from logged bandit feedback | IEEE Conference Publication | IEEE Xplore