Acquisition Functions in Bayesian Optimization

Bayesian optimization is effective in solving the optimization problem of black-box functions. In this work, the project focues on the optimization efficiency of three different acquisition functions (PI, EI and GP-LCB) based on the convergence speed of different test functions. At the beginning, we introduced the theorem of Bayesian optimization and gaussian process prior; later we showed the acquisition functions and benchmark functions we adopted; lastly, we carried out an experiment on the performance in different situations. In conclusion, mostly EI function is strongest but for some specific functions, PI and GP-LCB can be the more efficient one.