ArXiv Introduces PAWC Algorithm: Achieving O(N log N) Time Complexity for Partial Optimal Transport on the Circle

Summary:The ArXiv team introduces PAWC, a novel algorithm for solving the partial optimal transport problem on the circle. By preserving the line structure and introducing a free-gap invariant, PAWC reduces the computational complexity from O(N^2 log N) to O(N log N), while supporting cost calculations for all transported masses. Experimental results demonstrate PAWC's superior performance in handling periodic data such as angles, phases, and orientations, significantly improving computational efficienc

---

Background and Challenge

Partial Optimal Transport (POT) is a method for comparing two measures that allows for partial mass mismatch, making it robust to outliers, occlusions, and clutter. Traditionally, the computational complexity of POT problems is high, especially when dealing with periodic data such as angles, phases, and orientations, due to their global circulation characteristics.

Main Contributions

The ArXiv team introduces PAWC (Partial Optimal Transport on the Circle), a new algorithm designed to address the computational efficiency of partial optimal transport on the circle.

1. Algorithmic Innovation: PAWC preserves the line structure and introduces a free-gap invariant, avoiding the need for recalculating cuts at each step, thereby reducing the computational complexity from O(N^2 log N) to O(N log N).
2. Multi-scale Extension: The algorithm is not only applicable to one-dimensional circles but can also be extended to higher-dimensional spaces such as S^(d-1) by slicing over great circles.
3. Performance Advantage: At N=4096, PAWC computes the entire cost profile in 0.56 milliseconds, compared to 1.5 seconds for a single transported fraction from a general solver, demonstrating its significant efficiency gains in handling large-scale data.
4. Application Scenarios: PAWC retains 66% of the clean-data retrieval score on occluded, cluttered mpeg-7 shape images, while balanced circular OT retains only 16%. Additionally, on S^2, PAWC halves the fitting error of spherical sliced Wasserstein against contaminated targets, showcasing its advantages in handling complex geometric data.

Experiments and Results

Experimental results demonstrate that PAWC performs excellently in multiple benchmark tests, particularly in handling periodic data and complex geometric structures, significantly improving computational efficiency and accuracy.

Industry Impact and Recommendations

The introduction of PAWC provides an efficient new method for handling periodic data, with broad application prospects, including but not limited to:

• Computer Vision: PAWC can significantly improve computational efficiency when dealing with rotation-invariant data.
• Signal Processing: PAWC can provide more accurate transport calculations when handling periodic signals such as audio and radar data.
• Geographic Information Systems (GIS): PAWC can offer more efficient solutions when dealing with directional and angular data.

Developers can refer to the open-source code provided by ArXiv (https://github.com/mint-vu/Partial_Wasserstein_on_Circles) for further research and application development.

---

Source: ArXiv cs.LG (2026-08-24)

— END —

Tags: #ArXiv #POT #PAWC #Computational Geometry #Algorithm Optimization