Continuity equation on metric spaces via measure-valued derivations and BV-Wasserstein curves

What the math gives to ML

The paper gives a constructive representation of probability-measure dynamics, including discontinuous bounded-variation trajectories, through particle paths and measure-valued fluxes. The transferable asset is that Wasserstein variation charges both continuous transport and instantaneous jumps, while the weak continuity equation detects whether a distribution is being transported consistently rather than changed through untracked teleportation. A practical neural-network adaptation is a jump-aware regularizer for particle ensembles, latent samples, or mixture-of-experts routing states that permits resampling but explicitly charges the associated distributional displacement. This should improve training stability and make distribution shifts caused by discrete replacement operations measurable and controllable.