New Bayesian MF-LNO enhances active learning for complex PDEs

Researchers have developed a Bayesian multi-fidelity Laplace neural operator (MF-LNO) designed for active learning in oscillatory parametric partial differential equations (PDEs). This approach uses predictive uncertainty, quantified by replica-exchange stochastic gradient Langevin dynamics (reSGLD), to guide the acquisition of high-fidelity training data. Experiments on systems like the Lorenz system and Duffing oscillator show that this uncertainty-guided method is more data-efficient and accurate than random sampling and outperforms MF-DeepONet in predictive uncertainty quantification. AI

IMPACT This method offers a more data-efficient approach to modeling complex engineering systems, potentially accelerating design optimization and digital twin applications.

RANK_REASON The cluster contains an academic paper detailing a new method for solving PDEs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Bayesian Multi-Fidelity Laplace Neural Operator
• Haoyang Zheng
• Laplace Neural Operators
• Lorenz system
• MF-DeepONet
• replica-exchange stochastic gradient Langevin dynamics

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →