Search papers, labs, and topics across Lattice.
This paper introduces Displacement-Reshaped Optimal Transport (ReshapeOT), a method that improves the reliability of optimal transport for modeling distribution shifts by reshaping the ground metric using observed sample displacements. ReshapeOT replaces the Euclidean metric with a Mahalanobis distance estimated from displacement second moments, guiding the transport solutions. Experiments on synthetic and real-world data, along with practical use cases, demonstrate that ReshapeOT significantly enhances transport reliability.
Observed sample displacements can be integrated into optimal transport to carve expressways through the input space, leading to more reliable modeling of distribution shifts.
Optimal transport (OT) is a central framework for modeling distribution shifts. Because OT compares distributions directly in input space, a well-designed ground metric between observations is essential to ensure that the optimizer does not violate the true geometry of change. We propose Displacement-Reshaped Optimal Transport (ReshapeOT), a method that reshapes the ground metric by integrating observed sample displacements as an additional source of knowledge. Technically, ReshapeOT replaces the Euclidean metric with a Mahalanobis distance estimated from displacement second moments. This effectively carves expressways through the input space, inviting transport solutions that better align with observed displacements. Our method is computationally lightweight, integrates seamlessly into any OT solver that operates on a cost matrix, and can be kernelized for further flexibility. Experiments on synthetic and real-world data show that ReshapeOT achieves substantial gains in transport reliability. We further demonstrate our method's usefulness in two practical use cases.
