Learning with deep cascades

Giulia DeSalvo, Mehryar Mohri, Umar Syed

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

We introduce a broad learning model formed by cascades of predictors, Deep Cascades, that is structured as general decision trees in which leaf predictors or node questions may be members of rich function families. We present new data-dependent theoretical guarantees for learning with Deep Cascades with complex leaf predictors and node questions in terms of the Rademacher complexities of the sub-families composing these sets of predictors and the fraction of sample points reaching each leaf that are correctly classified. These guarantees can guide the design of a variety of different algorithms for deep cascade models and we give a detailed description of two such algorithms. Our second algorithm uses as node and leaf classifiers SVM predictors and we report the results of experiments comparing its performance with that of SVM combined with polynomial kernels.

Original languageEnglish (US)
Title of host publicationAlgorithmic Learning Theory - 26th International Conference, ALT 2015
EditorsClaudio Gentile, Sandra Zilles, Kamalika Chaudhuri
PublisherSpringer Verlag
Pages254-269
Number of pages16
ISBN (Print)9783319244853
DOIs
StatePublished - 2015
Event26th International Conference on Algorithmic Learning Theory, ALT 2015 - Banff, Canada
Duration: Oct 4 2015Oct 6 2015

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume9355
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other26th International Conference on Algorithmic Learning Theory, ALT 2015
Country/TerritoryCanada
CityBanff
Period10/4/1510/6/15

Keywords

  • Decision trees
  • Learning theory
  • Supervised learning

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science

Fingerprint

Dive into the research topics of 'Learning with deep cascades'. Together they form a unique fingerprint.

Cite this