SinkSLOT method offers faster optimal transport for large datasets

Researchers have introduced SinkSLOT, a novel method for entropic optimal transport (EOT) that significantly improves computational efficiency for large datasets. Unlike the standard Sinkhorn-Knopp algorithm, which requires O(N^2) operations per iteration, SinkSLOT reduces this to O(LN) with L slices. This advancement is achieved by using the expected sliced lifted transport plan to sparsify the Gibbs kernel, leading to substantial speedups over existing EOT methods. The proposed divergence also requires no debiasing and has demonstrated applicability in gradient flow experiments. AI

IMPACT Accelerates large-scale machine learning computations by improving the efficiency of optimal transport algorithms.

RANK_REASON The cluster describes a new computational method presented in an academic paper on arXiv. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• CatalyzeX
• DagsHub
• entropic optimal transport
• Gibbs kernel
• Hugging Face
• IArxiv
• optimal transport
• Sinkhorn-Knopp algorithm
• SinkSLOT

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →