GP-UCB
An acquisition function in Bayesian optimization that balances exploration and exploitation by adding the posterior mean to a confidence coefficient multiplied by the posterior standard deviation. The coefficient controls the strength of exploration.
Timeline (last 90 days)
Standard Bayesian optimization acquisition functions (such as GP-UCB, EI, Thompson sampling) are almost all designed to minimize cumulative regret, and therefore naturally favor exploitation over exploration.
The cumulative regret upper bound of standard GP-UCB is parameterized by the RKHS norm bound and the maximum information gain.
All Facts (2)
Standard Bayesian optimization acquisition functions (such as GP-UCB, EI, Thompson sampling) are almost all designed to minimize cumulative regret, and therefore naturally favor exploitation over exploration.
50%UnverifiedThe cumulative regret upper bound of standard GP-UCB is parameterized by the RKHS norm bound and the maximum information gain.
50%