Displacement-Sparse Neural Optimal Transport
Peter L. Chen, Yue Xie, Qingpeng Zhang, Department of Mathematics, Columbia University, HKU Musketeers Foundation Institute of Data Science, HKU-Shanghai Intelligent Computing Research Center · 2025
Read Displacement-Sparse Neural Optimal Transport by Peter L. Chen, Yue Xie, Qingpeng Zhang, Department of Mathematics, Columbia University, HKU Musketeers Foundation Institute of Data Science, HKU-Shanghai Intelligent Computing Research Center (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-01889.txt.
Description
Peter Chen, Yue Xie, and Qingpeng Zhang teach optimal transport maps that move mass with few nonzero shifts, making them readable in high dimensions, in work from twenty twenty-five. Neural transport solvers beat classic ones on speed and memory for tasks like drug response modeling. Their dense maps, though, are hard to interpret.
Earlier sparsity tricks worked only for exact solvers. Their new formulation brings displacement sparsity into neural solvers, with a smoothed counting penalty that beats the usual absolute value version. A flexible network design lets sparsity strength adapt during training. On large, high dimensional runs, the adaptive setup improved accuracy and stayed practical for downstream use.
Catalog record
| CREATOR | Peter L. Chen, Yue Xie, Qingpeng Zhang, Department of Mathematics, Columbia University, HKU Musketeers Foundation Institute of Data Science, HKU-Shanghai Intelligent Computing Research Center |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 72 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-01889