SLA learning from past failures, a Multi-Armed Bandit approach | IEEE Conference Publication | IEEE Xplore