Reproduce “Goal-Driven Adaptive Sampling Strategies for Machine Learning Models Predicting Fields” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jigar Parekh, Philipp Bekemeyer
Researchers at TU Braunschweig and DLR developed goal-driven adaptive sampling strategies for machine learning models that predict spatially distributed fields in engineering applications. Their unified framework, using coupled infill criteria, significantly improves the accuracy of both scalar performance metrics and high-dimensional field outputs, reducing the required number of high-fidelity simulations while precisely capturing complex flow features.
Key insights
- Coupling a robust scalar uncertainty estimator (GP) with a high-fidelity field predictor (GNN) through their disagreement (misfit or JSD) effectively guides adaptive sampling to improve both models simultaneously.
- Traditional scalar-only adaptive sampling is insufficient for adequately reducing uncertainty and improving accuracy in localized regions of high-dimensional field predictions.
- The proposed coupled strategies effectively reduce epistemic uncertainty across both scalar and spatial field outputs, leading to more reliable and accurate surrogate models for complex engineering problems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.