Reinforcement Learning-based Adaptive Mitigation of Uncorrected DRAM Errors in the Field.