Enabling high-dimensional surrogate-assisted optimization by using sliding windows

A Review of Population-Based Metaheuristics for Large-Scale Black-Box Global Optimization—Part II

“…However, exploiting the structure is a more challenging task as compared to partially separable problems. Despite the importance of overlapping problems, very few works have been dedicated to large-scale overlapping problems [78,123,124]. This section reviews some of such techniques that can help with the scalability of algorithms for large-scale global optimization.…”

“…Werth et al [78] used a sliding window mechanism and optimizes the variables inside the window to reduce the dimension of the problem. The problem structure is taken into account by iteratively constructing the interaction matrix of the problem using LINC-R; however, instead of finding the entire matrix, only the interactions within a given sliding window are considered at each iteration.…”

“…Sun et al [214] extended the CEC'2013 suite to make it more tunable. Other studies which paid attention to problems with overlapping components are conducted by Sayed et al [215] and Werth et al [78].…”

Decomposition for Large-scale Optimization Problems with Overlapping Components

“…Apart from the above methods that decompose decision variables into mutually exclusive subsets, there have been other techniques that partition a large-scale problem into overlapping sub-problems [17]- [22]. However it will raise another challenge; that is how to exchange information for a shared variable between overlapping components.…”