An O(n log n) Cutting Plane Algorithm for Structured Output Ranking - Inria - Institut national de recherche en sciences et technologies du numérique Access content directly
Conference Papers Year : 2014

An O(n log n) Cutting Plane Algorithm for Structured Output Ranking

Abstract

In this work, we consider ranking as a training strategy for structured output prediction. Recent work has begun to explore structured output prediction in the ranking setting, but has mostly focused on the special case of bipartite preference graphs. The bipartite special case is computationally efficient as there exists a linear time cutting plane training strategy for hinge loss bounded regularized risk, but it is unclear how to feasibly extend the approach to complete preference graphs. We develop here a highly parallelizable O(n log n) algorithm for cutting plane training with complete preference graphs that is scalable to millions of samples on a single core. We explore theoretically and empirically the relationship between slack rescaling and margin rescaling variants of the hinge loss bound to structured losses, showing that the slack rescaling variant has better stability properties and empirical performance with no additional computational cost per cutting plane iteration. We further show generalization bounds based on uniform convergence. Finally, we demonstrate the effectiveness of the proposed family of approaches on the problem of object detection in computer vision.
Fichier principal
Vignette du fichier
StructuredOutputRanking.pdf (5.45 Mo) Télécharger le fichier
Origin : Files produced by the author(s)
Loading...

Dates and versions

hal-01020943 , version 1 (15-07-2014)

Identifiers

Cite

Matthew Blaschko, Arpit Mittal, Esa Rahtu. An O(n log n) Cutting Plane Algorithm for Structured Output Ranking. German Conference on Pattern Recognition, Sep 2014, Münster, Germany. ⟨10.1007/978-3-319-11752-2_11⟩. ⟨hal-01020943⟩
228 View
363 Download

Altmetric

Share

Gmail Facebook Twitter LinkedIn More