Distributed Bayesian Optimization

dbo is a compact python package for distributed bayesian optimization in multi-agent systems such as robotics source seeking problems. dbo utilizes sklearn GaussianProcessRegressor to model the surrogate function and offers multiple strategies to select queries. In addition to the vanilla bayesian optimization algorithm, dbo offers:

• Distributed and parallel optimization
• Stochastic policy evaluations
• Expected acquisition policy over pending queries
• Internal acquisition function regularization
• Regret analysis outputs

The research was conducted at Harvard University.