Search papers, labs, and topics across Lattice.
This paper introduces a scalable Sinkhorn coreset method that leverages non-uniform weights for entropically regularized Optimal Transport (OT) coupling, enabling closed-form updates and enhancing computational efficiency. The method generalizes k-means through soft assignments and is shown to maintain asymptotic consistency and Lipschitz stability against data perturbations. Empirical results demonstrate that the proposed approach achieves competitive or superior approximation quality while significantly reducing runtime, particularly in large-scale scenarios.
Achieving competitive approximation quality with a coreset method that cuts runtime dramatically, especially for large datasets, could redefine efficiency in data selection for machine learning.
Coresets distill large datasets into small, representative subsets for efficient downstream learning. Yet Optimal Transport (OT)-based selection typically requires intensive computation of transport plans, limiting scalability. We introduce a scalable Sinkhorn coreset method that permits closed-form updates of the entropically regularized OT coupling by allowing non-uniform coreset weights. This produces centroids that generalize k-means via soft assignments. We establish asymptotic consistency of the selected measure and Lipschitz stability to data perturbations, providing accuracy and robustness guarantees. Across synthetic and real-world benchmarks, the proposed method achieves competitive or improved approximation quality while substantially reducing runtime compared to Wasserstein- and standard Sinkhorn-based coreset selection, especially at large scale.
