CORR
2012
Springer
12 years 8 months ago
2012
Springer
We present a bandit algorithm, SAO (Stochastic and Adversarial Optimal), whose regret is, essentially, optimal both for adversarial rewards and for stochastic rewards. Specifical...