Skip to main navigation Skip to search Skip to main content

Enhanced Transformer Powered Time-Series Anomaly Detection via Contrastive Embedding and Unified Discrimination

  • Shuhan Wang
  • , Kaizhe Wan
  • , Xiaoyu Yu
  • , Muyan Yao
  • , Dan Tao*
  • *Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

This paper presents an enhanced Anomaly Transformer for unsupervised anomaly detection in multivariate time series. We introduce a contrastive embedding module that refines the input embedding by optimizing a lightweight MLP with an InfoNCE objective, where positive pairs are pulled closer while negative pairs are pushed apart in the feature space. We further propose Unified Discrimination at the decision stage, which unifies reconstruction error and association discrepancy into a single anomaly score. Experiments on four large-scale benchmarks (SMD, PSM, MSL, and SMAP) demonstrate consistent performance gains over the Anomaly Transformer baseline, improving Precision and F1-score while maintaining high Recall, with an average F1-score of 0.9554.

Original languageEnglish
Title of host publication2026 IEEE International Conference on Consumer Electronics - Taiwan (ICCE-Taiwan)
PublisherInstitute of Electrical and Electronics Engineers (IEEE)
Pages821-822
Number of pages2
ISBN (Electronic)9798319506429
ISBN (Print)9798319506436 (PoD)
DOIs
Publication statusPublished - 24 Aug 2026
Event13th IEEE International Conference on Consumer Electronics - Taiwan - Taoyuan, Taiwan, Province of China
Duration: 1 Jul 2026 → 3 Jul 2026

Publication series

NameIEEE International Conference on Consumer Electronics-Taiwan
PublisherIEEE
ISSN (Print)2381-5779

Conference

Conference13th IEEE International Conference on Consumer Electronics - Taiwan
Abbreviated titleICCE-Taiwan 2026
Country/TerritoryTaiwan, Province of China
CityTaoyuan
Period1/07/26 → 3/07/26

Bibliographical note

Publisher Copyright:
© 2026 IEEE.

Keywords

  • anomaly detection
  • contrastive embedding
  • Time-series data
  • Transformer
  • Unified Discrimination

ASJC Scopus subject areas

  • Human-Computer Interaction
  • Electrical and Electronic Engineering
  • Media Technology
  • Modelling and Simulation
  • Instrumentation

Fingerprint

Dive into the research topics of 'Enhanced Transformer Powered Time-Series Anomaly Detection via Contrastive Embedding and Unified Discrimination'. Together they form a unique fingerprint.

Cite this