New framework AD-RaNN optimizes randomized neural networks for PDEs

Researchers have introduced Adaptive-Distribution Randomized Neural Networks (AD-RaNN), a new framework designed to improve the performance of randomized neural networks in solving partial differential equations (PDEs). This method addresses the sensitivity of existing models to the sampling distribution of hidden-layer parameters by optimizing a low-dimensional vector that defines this distribution. AD-RaNN employs a two-stage training process, incorporating adaptive mechanisms like PDE-Driven Adaptive Distribution (PDAD) and Data-Driven Adaptive Distribution (DDAD) to enhance accuracy and reduce manual tuning. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Introduces a new method for optimizing neural network parameters in PDE solvers, potentially improving accuracy and reducing manual tuning.

RANK_REASON This is a research paper introducing a novel framework for solving PDEs.

Read on arXiv cs.LG →

• Adaptive-Distribution Randomized Neural Networks
• AD-RaNN
• Randomized neural networks
• RaNNs
• PDEs
• PDAD
• DDAD

• paper
• other