The RBF Hyperparameter in Evolutionary Surrogate-assisted Optimization

פרסום מחקרי: פרק בספר / בדוח / בכנספרסום בספר כנסביקורת עמיתים

תקציר

Expensive 'black-box' optimization problems involve functions whose analytic expression is unavailable and which are computationally-intensive to evaluate. Such problems are often solved with frameworks which combine evolutionary algorithms and surrogate models. A common surrogate variant is the Radial Basis Functions model which combines the output of several basis functions. The latter depend on a hyperparameter which affects their features and as such it also affects the overall surrogate prediction. Typically the optimal hyperparameter value is unknown and is therefore estimated by additional numerical procedures. This raises the question if the additional computational resources spent on the hyperparameter calibration are justified, namely do they translate to meaningful differences in the overall search effectiveness of the algorithm. This aspect has largely not been examined in the literature and typically the hyperparameter impact was assessed based only on the prediction accuracy of the stand-alone surrogate model. As such this paper addresses this open question and analyzes the impact of the hyperparameter on the overall search performance based on an extensive set of numerical experiments. A detailed analysis shows that modifying the hyperparameter strongly affected this performance and that the extent of the impact was related to the basis function type, the objective function modality, and the problem dimension.

שפה מקוריתאנגלית
כותר פרסום המארחProceedings - 2022 International Conference on Machine Learning, Control, and Robotics, MLCR 2022
מוציא לאורInstitute of Electrical and Electronics Engineers Inc.
עמודים38-42
מספר עמודים5
מסת"ב (אלקטרוני)9781665454599
מזהי עצם דיגיטלי (DOIs)
סטטוס פרסוםפורסם - 2022
אירוע2022 International Conference on Machine Learning, Control, and Robotics, MLCR 2022 - Suzhou, סין
משך הזמן: 29 אוק׳ 202231 אוק׳ 2022

סדרות פרסומים

שםProceedings - 2022 International Conference on Machine Learning, Control, and Robotics, MLCR 2022

כנס

כנס2022 International Conference on Machine Learning, Control, and Robotics, MLCR 2022
מדינה/אזורסין
עירSuzhou
תקופה29/10/2231/10/22

טביעת אצבע

להלן מוצגים תחומי המחקר של הפרסום 'The RBF Hyperparameter in Evolutionary Surrogate-assisted Optimization'. יחד הם יוצרים טביעת אצבע ייחודית.

פורמט ציטוט ביבליוגרפי