![[2602.15393] Doubly Stochastic Mean-Shift Clustering](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.15393] Doubly Stochastic Mean-Shift Clustering
Summary
The paper presents Doubly Stochastic Mean-Shift (DSMS), an innovative clustering algorithm that enhances standard Mean-Shift methods by introducing randomness in both trajectory updates and kernel bandwidth, improving stability and performance in sparse data scenarios.
Why It Matters
This research addresses the limitations of traditional Mean-Shift algorithms, particularly in data-scarce environments where over-segmentation occurs. By proposing a new approach that incorporates randomness, it offers a more robust solution for clustering in machine learning, which is crucial for various applications in data science and computer vision.
Key Takeaways
- DSMS introduces randomness in both trajectory updates and kernel bandwidth.
- The method significantly outperforms traditional Mean-Shift algorithms in sparse clustering scenarios.
- Randomized bandwidth acts as an implicit regularization mechanism.
- The approach prevents over-segmentation without degrading performance.
- Theoretical convergence results support the effectiveness of DSMS.
Computer Science > Machine Learning arXiv:2602.15393 (cs) [Submitted on 17 Feb 2026] Title:Doubly Stochastic Mean-Shift Clustering Authors:Tom Trigano, Yann Sepulcre, Itshak Lapidot View a PDF of the paper titled Doubly Stochastic Mean-Shift Clustering, by Tom Trigano and 1 other authors View PDF HTML (experimental) Abstract:Standard Mean-Shift algorithms are notoriously sensitive to the bandwidth hyperparameter, particularly in data-scarce regimes where fixed-scale density estimation leads to fragmentation and spurious modes. In this paper, we propose Doubly Stochastic Mean-Shift (DSMS), a novel extension that introduces randomness not only in the trajectory updates but also in the kernel bandwidth itself. By drawing both the data samples and the radius from a continuous uniform distribution at each iteration, DSMS effectively performs a better exploration of the density landscape. We show that this randomized bandwidth policy acts as an implicit regularization mechanism, and provide convergence theoretical results. Comparative experiments on synthetic Gaussian mixtures reveal that DSMS significantly outperforms standard and stochastic Mean-Shift baselines, exhibiting remarkable stability and preventing over-segmentation in sparse clustering scenarios without other performance degradation. Comments: Subjects: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV) Cite as: arXiv:2602.15393 [cs.LG] (or arXiv:2602.15393v1 [cs.LG] for this version) http...