Deep metric learning for sequential data using approximate information

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademicpeer-review

1 Citation (Scopus)
2 Downloads (Pure)

Abstract

Learning a distance metric provides solutions to many problems where the data exists in a high dimensional space and hand-crafted distance metrics fail to capture its semantical structure. Methods based on deep neural networks such as Siamese or Triplet networks have been developed for learning such metrics. In this paper we present a metric learning method for sequence data based on a RNN-based triplet network. We posit that this model can be trained efficiently with regards to labels by using Jaccard distance as a proxy distance metric. We empirically demonstrate the performance and efficiency of the approach on three different computer log-line datasets.

Original languageEnglish
Title of host publicationMachine Learning and Data Mining in Pattern Recognition - 14th International Conference, MLDM 2018, Proceedings
EditorsPetra Perner
Place of PublicationCham
PublisherSpringer
Pages269-282
Number of pages14
ISBN (Electronic)978-3-319-96136-1
ISBN (Print)978-3-319-96135-4
DOIs
Publication statusPublished - 8 Jul 2018
Event14th International Conference on Machine Learning and Data Mining in Pattern Recognition, MLDM 2018 - New York, United States
Duration: 15 Jul 201819 Jul 2018

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume10934 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference14th International Conference on Machine Learning and Data Mining in Pattern Recognition, MLDM 2018
Country/TerritoryUnited States
CityNew York
Period15/07/1819/07/18

Keywords

  • Deep learning
  • Efficient metric learning
  • Triplet network

Fingerprint

Dive into the research topics of 'Deep metric learning for sequential data using approximate information'. Together they form a unique fingerprint.

Cite this