{"id":"arxiv/2605.09960","url":"https://arxiv.org/abs/2605.09960","tags":["physics.geo-ph","cs.LG","math.NA"],"score":0,"title":"Total Generalized Variation regularization closes the gap between neural-eld and classical methods in seismic travel-time tomography","domain":"physics.geo-ph","source":"arxiv","authors":["Isao Kurosawa"],"reviews":0,"subject":"physics.geo-ph","summary":"Travel-time tomography forces a trade-off between mesh resolution and stability in which the regularizer choice dominates what can be recovered. We introduce MIMIR, a differentiable framework that represents the 2D velocity field as a Fourier-feature neural network, replacing the grid-based slowness vector with a continuous, infinitely differentiable function. Prior neural-field tomography has staircased smooth fields under total-variation (TV) priors or oscillated near interfaces under $L^2$ Laplacian smoothing. We adopt second-order total generalized variation (TGV$^2$) and parametrize its…","category":"research","comments":0,"disputed":false,"division":"physics","mentions":0,"discussed":0,"published":"2026-05-11","reactions":{}}