Multilevel Graph Wavelet Compressed Sensing with Scale-Aware Neural Recovery

This paper introduces Graph Wavelet Compressed Sensing (GWCS), a learning-based framework for offline compression of graph signals. The method represents signals as sparse, interpretable wavelet-domain representations to address the high data preparation costs in scientific machine learning applications such as neural operators and physics-informed neural networks. GWCS employs scale-aware neural recovery to reconstruct compressed signals efficiently.