Abstract:Many real-world engineering and industrial optimization problems involve expensive function evaluations (e.g., computer simulations and physical experiments) and possess a large number of decision variables. In many such scenarios, the optimization task has to be performed based on the previously available simulation data only. Also, to cut down the experimental expenses, it has been an open-ended research area to approximate these expensive function evaluations using a less expensive data-driven model trained on historical offline data collected from various expensive simulations. Offline data-driven evolutionary algorithms (DDEAs) efficiently use available data and surrogates to guide the optimization process. However, while building the surrogate models from limited offline data, the existing methods lack properties like reliability, scalability, and robustness. To address these challenges, in this work, we have proposed novel active learning with a reliability sampling-based evolutionary framework, where an ensemble of heterogeneous RBFNs (radial basis function neural networks), acting as an active learner, can gain insightful knowledge from unlabeled data. Consequently, a novel reliability sampling for selecting query individuals is also introduced for model management. It selects the utmost reliable candidate solutions for which the ensemble members have the least conflict to enrich the training data. These reliable query individuals are labeled as a weighted sum of RBFNs output, where the RBFN model with less error has more weightage. Furthermore, this article discusses the theoretical idea from which reliability sampling-based query strategy is inspired and provides the complexity analysis. Moreover, the results on five benchmark problems and CES 2017 test suit with 10, 30, 50, and 100-dimensional decision variables show that the proposed algorithm has achieved highly competitive results compared with four state-of-the-art offline DDEAs and three online DDEAs. Finally, our algorithm is applied to an offline data-driven expensive engineering optimization problem-aerodynamic airfoil design optimization from offline data to verify its efficacy on real-world problems. The experimental results demonstrate that the proposed active learning-based framework provides a highly reliable, efficient, and robust surrogate to assist evolutionary algorithms in solving offline data-driven expensive engineering problems.

Data efficient surrogate modeling for engineering design: Ensemble-free batch mode deep active learning for regression

Deep Active Learning for Regression Using $ε$-weighted Hybrid Query Strategy

Deep Residual Surrogate Model

Active-Learning-Driven Surrogate Modeling for Efficient Simulation of Parametric Nonlinear Systems

ALeRSa-DDEA: active learning with reliability sampling-based evolutionary algorithm framework for solving offline data-driven expensive engineering optimization

Deep Learning-based Finite Element Analysis (FEA) surrogate for sub-sea pressure vessel

Physics-Based Active Learning for Design Space Exploration and Surrogate Construction for Multiparametric Optimization

Feasibility Study on Active Learning of Smart Surrogates for Scientific Simulations

Active learning for regression in engineering populations: A risk-informed approach

Active Learning for Neural PDE Solvers

Airfoil Aerodynamic Optimization Design Using Ensemble Learning Surrogate Model

A new approach for data-driven surrogate modelling applied in highly nonlinear engineering functions

A Least Squares Ensemble Model Based on Regularization and Augmentation Strategy

Two-Stage Surrogate Modeling for Data-Driven Design Optimization with Application to Composite Microstructure Generation

A novel data generation scheme for surrogate modelling with deep operator networks

MelissaDL x Breed: Towards Data-Efficient On-line Supervised Training of Multi-parametric Surrogates with Active Learning

Deep Bayesian surrogate models with adaptive online sampling for ensemble-based data assimilation

Enhanced data efficiency using deep neural networks and Gaussian processes for aerodynamic design optimization

Active Learning over DNN: Automated Engineering Design Optimization for Fluid Dynamics Based on Self-Simulated Dataset

Active learning of deep surrogates for PDEs: application to metasurface design

A Fast Learning-Based Surrogate of Electrical Machines using a Reduced Basis