Adaptive objective selection for multi-fidelity optimization
| Autor: | Takahiro Yamaguchi, Youhei Akimoto, Takuma Shimizu |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
Mathematical optimization
010201 computation theory & mathematics Computer science media_common.quotation_subject 0202 electrical engineering electronic engineering information engineering Evolutionary algorithm Fidelity 020201 artificial intelligence & image processing 0102 computer and information sciences 02 engineering and technology 01 natural sciences Selection (genetic algorithm) media_common |
| Zdroj: | GECCO |
| DOI: | 10.1145/3321707.3321709 |
| Popis: | In simulation-based optimization we often have access to multiple simulators or surrogate models that approximate a computationally expensive or intractable objective function with different tradeoffs between the fidelity and computational time. Such a setting is called multi-fidelity optimization. In this paper, we propose a novel strategy to adaptively select which simulator to use during optimization of comparison-based evolutionary algorithms. Our adaptive switching strategy works as a wrapper of multiple simulators: optimization algorithms optimize the wrapper function and the adaptive switching strategy selects a simulator inside the wrapper, implying wide applicability of the proposed approach. We empirically investigate how efficiently the adaptive switching strategy manages simulator selection on test problems and theoretically investigate how it changes the fidelity level during optimization. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|