Well-posedness for the mean curvature flow on the half-space and on bounded domains

What the math gives to ML

The paper provides a slope-dependent anisotropic diffusion law for vector-valued graph outputs. Diffusion is controlled by the induced graph metric, so the layer smooths curvature without applying isotropic blur indiscriminately. The most practical transfer is a differentiable mean-curvature-flow relaxation layer after a network predicts a spatial vector field, with explicit boundary handling and slope monitoring for stability.