Researchers have developed a novel method for training neural network surrogates that can significantly accelerate simulations of wave-scattering inverse problems. This new approach overcomes previous scaling limitations by dynamically generating salient training examples during the training process, rather than relying on random sampling. The technique has successfully trained a surrogate for two-dimensional wave scattering with over 41,000 variables, a more than 73-fold increase, and demonstrated its effectiveness in designing photonic components like beam splitters and gradient-index lenses. AI
IMPACT This research could enable significantly faster and more scalable simulations for designing optical components and other wave-scattering problems.
RANK_REASON Academic paper detailing a new method for training neural networks for scientific simulation. [lever_c_demoted from research: ic=1 ai=1.0]
- artificial neural network
- arXiv
- beam splitters
- finite-difference time-domain method
- photonic crystal
- wave-scattering inverse problems
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
