{"id":"arxiv/2606.16913","url":"https://arxiv.org/abs/2606.16913","tags":["math.ST","math.NA","stat.ML"],"score":0,"title":"Optimal Multiscale Learning of Linear Operators","domain":"math.ST","source":"arxiv","authors":["Jiaheng Chen","Daniel Sanz-Alonso"],"reviews":0,"subject":"math.ST","summary":"We study the statistical and computational limits of learning bounded linear operators between Sobolev spaces from noisy input-output data. In wavelet coordinates, the problem is recast as an infinite-dimensional matrix regression problem with a heterogeneous two-sided multiscale structure. We establish minimax rates under Sobolev operator-norm loss and construct a finite-resolution blockwise least-squares estimator attaining these rates. The analysis reveals a nonuniform local estimation difficulty across scales, which can be exploited algorithmically: by assigning scale-adaptive sample…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-06-15","reactions":{}}