雅可比增强神经网络

Jun, 2024

Jacobian-Enhanced Neural Networks

Steven H. Berguin

TL;DRJacobian-Enhanced Neural Networks (JENN) are used in computer-aided design to accurately predict partial derivatives and provide a faster alternative to physics-based models known as surrogate models or meta-models, particularly for surrogate-based optimization.

Abstract

jacobian-enhanced neural networks (JENN) are densely connected multi-layer perceptrons, whose training process is modified to predict partial derivatives accurately. Their main benefit is better accuracy with few