AdaE-SAEA: New MetaBBO method balances robustness and accuracy in optimization
This adaptive algorithm uses ensemble surrogate modeling to dynamically trade off exploration and exploitation...
Surrogate-assisted evolutionary algorithms (SAEAs) are widely used for expensive black-box optimization, but they rely on rigid, manually designed components, limiting flexibility and generalization. Meta-black-box optimization (MetaBBO) offers adaptive component control, yet existing methods typically control only a single component and rarely consider the critical robustness–accuracy trade-off in surrogate modeling—vital for stable early exploration and precise late exploitation.
To address this, Xiao Jin and colleagues propose AdaE-SAEA, which embeds SAEA as a low-level optimizer within the MetaBBO framework and jointly controls the infill criterion and ensemble surrogate modeling. Specifically, it uses bagging and boosting as surrogate modeling modules to adaptively balance robustness and accuracy across different search phases, while the meta-policy simultaneously selects the infill criterion for adaptive sampling. The meta-policy is trained via reinforcement learning with parallel sampling and centralized training, improving both efficiency and transferability. Experiments on synthetic and real-world problems show AdaE-SAEA outperforms state-of-the-art baselines and MetaBBO methods, and it demonstrates the effectiveness of TabPFN as a base surrogate model. This is the first work to unify the control of surrogate modeling and infill criteria in SAEAs while explicitly addressing the robustness–accuracy trade-off.
- AdaE-SAEA combines ensemble surrogate modeling (bagging and boosting) to adaptively balance robustness and accuracy across optimization phases.
- Meta-policy trained via reinforcement learning with parallel sampling and centralized training controls both infill criterion and surrogate model selection.
- Outperforms state-of-the-art SAEA and MetaBBO baselines on synthetic and real-world expensive optimization problems.
Why It Matters
Makes expensive optimization (e.g., engineering design, hyperparameter tuning) more adaptive and efficient by automating the trade-off between exploration and exploitation.