This paper addresses continuous global optimization problems in which the objective function cannot be evaluated directly. Instead, the optimizer observes only which of two selected design points appears to perform better, and this observation is subject to noise. Such a setting is common when feedback comes from human preference or noisy comparisons.

The work centers on a fully sequential Gaussian process regression approach for preference learning. In this framework, a Gaussian process serves as a surrogate model of the unknown objective, and the 'fully sequential' aspect indicates that the model is updated after each pairwise comparison, with the next pair chosen based on the current posterior.

The provided abstract is truncated, so specific algorithmic details, acquisition strategies, or experimental results are not included in the source. The contribution, as stated, is the study of this problem class and the development of a sequential GP-based method for it.