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
PublisherSpringer Verlag
Pages254-269
Number of pages16
Volume9355
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)03029743
ISSN (Electronic)16113349

Other

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

Fingerprint

Cascade
Predictors
Leaves
Decision trees
Vertex of a graph
Classifiers
Polynomials
Sample point
Dependent Data
Decision tree
Classifier
Learning
kernel
Experiments
Polynomial
Model
Experiment
Family

Keywords

  • Decision trees
  • Learning theory
  • Supervised learning

ASJC Scopus subject areas

  • Computer Science(all)
  • Theoretical Computer Science

Cite this

DeSalvo, G., Mohri, M., & Syed, U. (2015). Learning with deep cascades. In Algorithmic Learning Theory - 26th International Conference, ALT 2015 (Vol. 9355, pp. 254-269). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 9355). Springer Verlag. https://doi.org/10.1007/978-3-319-24486-0_17

Learning with deep cascades. / DeSalvo, Giulia; Mohri, Mehryar; Syed, Umar.

Algorithmic Learning Theory - 26th International Conference, ALT 2015. Vol. 9355 Springer Verlag, 2015. p. 254-269 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 9355).

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

DeSalvo, G, Mohri, M & Syed, U 2015, Learning with deep cascades. in Algorithmic Learning Theory - 26th International Conference, ALT 2015. vol. 9355, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 9355, Springer Verlag, pp. 254-269, 26th International Conference on Algorithmic Learning Theory, ALT 2015, Banff, Canada, 10/4/15. https://doi.org/10.1007/978-3-319-24486-0_17
DeSalvo G, Mohri M, Syed U. Learning with deep cascades. In Algorithmic Learning Theory - 26th International Conference, ALT 2015. Vol. 9355. Springer Verlag. 2015. p. 254-269. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-319-24486-0_17
DeSalvo, Giulia ; Mohri, Mehryar ; Syed, Umar. / Learning with deep cascades. Algorithmic Learning Theory - 26th International Conference, ALT 2015. Vol. 9355 Springer Verlag, 2015. pp. 254-269 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{c402219608a949e2a4c3c624a2f034d5,
title = "Learning with deep cascades",
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.",
keywords = "Decision trees, Learning theory, Supervised learning",
author = "Giulia DeSalvo and Mehryar Mohri and Umar Syed",
year = "2015",
doi = "10.1007/978-3-319-24486-0_17",
language = "English (US)",
isbn = "9783319244853",
volume = "9355",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "254--269",
booktitle = "Algorithmic Learning Theory - 26th International Conference, ALT 2015",

}

TY - GEN

T1 - Learning with deep cascades

AU - DeSalvo, Giulia

AU - Mohri, Mehryar

AU - Syed, Umar

PY - 2015

Y1 - 2015

N2 - 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.

AB - 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.

KW - Decision trees

KW - Learning theory

KW - Supervised learning

UR - http://www.scopus.com/inward/record.url?scp=84945954110&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84945954110&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-24486-0_17

DO - 10.1007/978-3-319-24486-0_17

M3 - Conference contribution

AN - SCOPUS:84945954110

SN - 9783319244853

VL - 9355

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 254

EP - 269

BT - Algorithmic Learning Theory - 26th International Conference, ALT 2015

PB - Springer Verlag

ER -