We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

stat.ML

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Statistics > Machine Learning

Title: Weston-Watkins Hinge Loss and Ordered Partitions

Abstract: Multiclass extensions of the support vector machine (SVM) have been formulated in a variety of ways. A recent empirical comparison of nine such formulations [Do\v{g}an et al. 2016] recommends the variant proposed by Weston and Watkins (WW), despite the fact that the WW-hinge loss is not calibrated with respect to the 0-1 loss. In this work we introduce a novel discrete loss function for multiclass classification, the ordered partition loss, and prove that the WW-hinge loss is calibrated with respect to this loss. We also argue that the ordered partition loss is maximally informative among discrete losses satisfying this property. Finally, we apply our theory to justify the empirical observation made by Do\v{g}an et al. that the WW-SVM can work well even under massive label noise, a challenging setting for multiclass SVMs.
Comments: 38 pages, 3 figures
Subjects: Machine Learning (stat.ML); Machine Learning (cs.LG); Optimization and Control (math.OC)
Cite as: arXiv:2006.07346 [stat.ML]
  (or arXiv:2006.07346v1 [stat.ML] for this version)

Submission history

From: Yutong Wang [view email]
[v1] Fri, 12 Jun 2020 17:34:21 GMT (62kb)

Link back to: arXiv, form interface, contact.