New SCORE method enhances physics-informed neural network training

Researchers have developed SCORE, a novel self-concordance-inspired quasi-Newton method designed to improve the training of physics-informed neural networks (PINNs). This method addresses challenges with indefinite and poorly scaled curvature in PINN objectives by using a quasi-Newton decrement to jointly determine candidate steps and adaptive shifts for secant geometry. Experiments on several partial differential equations, including the viscous Burgers' equation, demonstrate that SCORE achieves lower final errors compared to existing BFGS and self-scaled Broyden baselines. AI

IMPACT This new method could lead to more accurate and efficient solutions for complex scientific problems modeled by neural networks.

RANK_REASON The cluster contains a research paper detailing a new method for training neural networks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Broyden–Fletcher–Goldfarb–Shanno algorithm
• complex Ginzburg--Landau equation
• Korteweg--de Vries equation
• Kuramoto--Sivashinsky equation
• physics-informed neural networks
• viscous Burgers' equation

• paper
• infra

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