Towards self-adaptive efficient global optimization

. In this paper, it is proposed to dynamically control the trade-o ff between exploration and exploitation for the e ffi cient global optimization algorithm. To achieve this, we use the so-called Moment-Generating Function of Improvement criterion, in which an additional parameter, called “temperature”, is introduced to smoothly control the exploration / exploitation balance. It is proposed to adapt the temperature using two approaches: 1) a success-rate based mechanism 2) a self-adaptive algorithm where multiple di ff erent temperatures are used to generate new points. The temperature related to the best point is then selected for the next iteration. The self-adaptive algorithm is validated on two multi-modal functions and the result shows that the adaptive temperature mechanism speeds up the convergence rate of EGO, compared to the fixed temperature setting.