Quantifying the Impact of Lossy Compression on Neural Generative Surrogate Modeling
A new research paper explores the impact of lossy compression on neural generative surrogate models, which are used to approximate complex scientific simulations. The study quantifies how compression errors affect model quality, proposing a method to estimate the tolerable error level. This approach can significantly reduce data storage and training time while maintaining high-quality surrogate models, demonstrating savings of up to 39x in storage and 3x in training time. AI
IMPACT Enables significant reductions in data storage and training time for AI models used in scientific discovery.