CoChat

Self-Supervised Contrastive Representation Learning for Semi-Supervised Time-Series Classification

Open access
Eldele, Emadeldeen, Ragab, Mohamed, Chen, Zhenghua, Wu, Min, Kwoh, Chee-Keong, Li, Xiaoli, Guan, Cuntai · 2023 · IEEE Transactions on Pattern Analysis and Machine Intelligence

The DOI matches its Crossref record; 33 of the 54 references it cites were found in a registry or an index, and 21 are not checked yet; none of the 20 sources its claims cite is checked for support yet.

Abstract

Learning time-series representations when only unlabeled data or few labeled samples are available can be a challenging task. Recently, contrastive self-supervised learning has shown great improvement in extracting useful representations from unlabeled data via contrasting different augmented views of data. In this work, we propose a novel Time-Series representation learning framework via Temporal and Contextual Contrasting (TS-TCC) that learns representations from unlabeled data with contrastive learning. Specifically, we propose time-series-specific weak and strong augmentations and use their views to learn robust temporal relations in the proposed temporal contrasting module, besides learning discriminative representations by our proposed contextual contrasting module. Additionally, we conduct a systematic study of time-series data augmentation selection, which is a key part of contrastive learning. We also extend TS-TCC to the semi-supervised learning settings and propose a Class-Aware TS-TCC (CA-TCC) that benefits from the available few labeled data to further improve representations learned by TS-TCC. Specifically, we leverage the robust pseudo labels produced by TS-TCC to realize a class-aware contrastive loss. Extensive experiments show that the linear evaluation of the features learned by our proposed framework performs comparably with the fully supervised training. Additionally, our framework shows high efficiency in few labeled data and transfer learning scenarios.

Checked 2026-10-05 · 21 references not checked yet

Overview

CoChat Summary

TL;DR

The paper proposes TS-TCC, a self-supervised contrastive framework for time-series representation learning using weak/strong augmentations, temporal contrasting, and contextual contrasting, and extends it to semi-supervised CA-TCC using pseudo-labeled class-aware contrastive learning. Across ten real-world datasets, TS-TCC and CA-TCC outperform self-supervised and semi-supervised baselines under few-label and transfer-learning settings.

Key findings

  • With 1% labeled data, TS-TCC achieved the best average linear-evaluation accuracy across 10 datasets: 76.1% vs. 73.3% for CPC, 65.5% for SimCLR, 61.8% for supervised training, and 58.8% for random initialization; with 5% labels, TS-TCC averaged 80.6% vs. 80.0% for CPC and 71.8% for SimCLR.
  • CA-TCC’s average F1-score outperformed the second-best semi-supervised baseline by 3.8% and 4.9% with 1% and 5% labels, respectively; its average accuracy was 77.8% at 1% labels and 83.5% at 5% labels.
  • In cross-domain fault diagnosis transfer experiments, TS-TCC improved over supervised training by about 4% accuracy on average, and CA-TCC by about 6% accuracy on average.
  • Ablation showed the cross-view prediction task improved accuracy by more than 6% on HAR and about 2% on Sleep-EDF and Epilepsy; combining weak and strong augmentations gave the best HAR result at 77.58% accuracy with 5% labels.

Methods

  • Design: TS-TCC learns from unlabeled data via time-series-specific weak and strong augmentations, a temporal contrasting module with a tough cross-view prediction task, and a contextual contrasting module. CA-TCC extends this to semi-supervised learning in four phases: TS-TCC pretraining, fine-tuning with few labels, pseudo-label generation, and class-aware supervised contrastive training.
  • Data: ten public real-world time-series datasets were used, including UCI HAR, Sleep-EDF, Epilepsy, Fault Diagnosis, and seven UCR datasets (Wafer, FordA, FordB, POC, PPOC, StarLightCurves, ElectricDevices). Experiments used 60%/20%/20% train/validation/test splits, subject-wise split for Sleep-EDF, and 5 runs with different seeds.
  • Analysis: performance was measured by accuracy and macro-averaged F1-score, with linear evaluation on frozen pretrained encoders, comparisons against self-supervised and semi-supervised baselines, transfer-learning tests, ablation, sensitivity analysis, and augmentation-selection experiments.

Limitations

  • The paper does not include a dedicated limitations section. It states that time-series augmentation selection remains an open problem and that proper augmentation parameters vary by dataset characteristics, requiring normalization and dataset-specific choices.
  • The paper notes that ensembling multiple self-supervised algorithms gave only minor improvement while increasing complexity, training time, and computational demands.
  • Evident from the text: evaluation is limited to ten datasets and selected label budgets, sensitivity analysis is performed only on HAR, and CA-TCC depends on the quality of pseudo labels generated by the fine-tuned TS-TCC encoder.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1109/tpami.2023.3308189

Cites

33 of 54 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

0 Valid · 0 Not valid · 12 Not checked · 87 Own results and interpretations

59 more claims

Create a free account to see all 99 claims and to verify the 20 references still open.

See all claims — free account
  • It may state two separate claims as one. 24
  • Probably not a substantial claim. 14

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Research works printed but not linked: 3

  1. 10.5555/3524938.3525087 · doi:10.5555/3524938.3525087
  2. 10.5555/3495724.3497510 · doi:10.5555/3495724.3497510
  3. 10.1161/01.cir.101.23.e21 · doi:10.1161/01.cir.101.23.e21

Cited by

See all 224

Metadata

Authors
Eldele, E., Ragab, M., Chen, Z., Wu, M., Kwoh, C.-K., Li, X., & Guan, C.
Affiliations
  • Agency for Science, Technology and Research (SG)
  • Nanyang Technological University (SG)
Venue
IEEE Transactions on Pattern Analysis and Machine Intelligence
Published
2023-12
Volume
45
Issue
12
Pages
15604-15618
ISSN
  • 0162-8828 (ISSN-L)
  • 2160-9292
  • 1939-3539
OpenAlex
W4386212336
Cited by
224
References
51
Citations by year
2026: 71 · 2025: 89 · 2024: 55 · 2023: 9
FWCI
27.15
Type
journal article
Language
en
Open access
Locations
  • IEEE Transactions on Pattern Analysis and Machine Intelligence (publishedVersion)
  • PubMed (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • DR-NTU (Nanyang Technological University) (submittedVersion)
  • arXiv (Cornell University) (cc-by)
Funders
  • A*STAR SINGA Scholarship
  • Agency for Science, Technology and Research
  • AME Programmatic Funds (A20H6b0151)
  • Career Development (C210112046)
  • MOE Academic Research (MOE2019-T2-2-175)
Topics
  • Physical Sciences › Computer Science › Signal Processing › Time Series Analysis and Forecasting
  • Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
  • Physical Sciences › Computer Science › Signal Processing › Music and Audio Processing
Keywords
time series classification, semi-supervised learning, self-supervised learning, contrastive learning, data augmentation, pseudo-labeling, few labeled samples, transfer learning, representation learning

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-05, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-05 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.