Transfer Learning for Time Series Classification Using Synthetic Data Generation

Yarden Rotem, Nathaniel Shimoni, Lior Rokach, Bracha Shapira

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

Abstract

In this paper, we propose an innovative Transfer learning for Time series classification method. Instead of using an existing dataset from the UCR archive as the source dataset, we generated a 15,000,000 synthetic univariate time series dataset that was created using our unique synthetic time series generator algorithm which can generate data with diverse patterns and angles and different sequence lengths. Furthermore, instead of using classification tasks provided by the UCR archive as the source task as previous studies did, we used our own 55 regression tasks as the source tasks, which produced better results than selecting classification tasks from the UCR archive.

Original languageAmerican English
Title of host publicationCyber Security, Cryptology, and Machine Learning - 6th International Symposium, CSCML 2022, Proceedings
Subtitle of host publication6th International Symposium Be'er Sheva, Israel, CSCML
EditorsShlomi Dolev, Amnon Meisels, Jonathan Katz
PublisherSpringer Science and Business Media Deutschland GmbH
Pages232-246
Number of pages15
ISBN (Electronic)978-3-031-07689-3
ISBN (Print)978-3-031-07688-6
DOIs
StatePublished - 23 Jun 2022
Event6th International Symposium on Cyber Security Cryptography and Machine Learning, CSCML 2022 - Beer Sheva, Israel
Duration: 30 Jun 20221 Jul 2022

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13301 LNCS

Conference

Conference6th International Symposium on Cyber Security Cryptography and Machine Learning, CSCML 2022
Country/TerritoryIsrael
CityBeer Sheva
Period30/06/221/07/22

Keywords

  • Synthetic data
  • Time series classification
  • Transfer learning

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • General Computer Science

Fingerprint

Dive into the research topics of 'Transfer Learning for Time Series Classification Using Synthetic Data Generation'. Together they form a unique fingerprint.

Cite this