Contrastive Representation Learning: A Framework and Review
Open accessThe DOI matches its Crossref record; 147 of the 162 references it cites were found in a registry or an index, and 15 are not checked yet.
Abstract
Contrastive Learning has recently received interest due to its success in self-supervised representation learning in the computer vision domain. However, the origins of Contrastive Learning date as far back as the 1990s and its development has spanned across many fields and domains including Metric Learning and natural language processing. In this paper, we provide a comprehensive literature review and we propose a general Contrastive Representation Learning framework that simplifies and unifies many different contrastive learning methods. We also provide a taxonomy for each of the components of contrastive learning in order to summarise it and distinguish it from other forms of machine learning. We then discuss the inductive biases which are present in any contrastive learning system and we analyse our framework under different views from various sub-fields of Machine Learning. Examples of how contrastive learning has been applied in computer vision, natural language processing, audio processing, and others, as well as in Reinforcement Learning are also presented. Finally, we discuss the challenges and some of the most promising future research directions ahead.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
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?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“According to the suggestions of Le-Khac et al. [29], the main methods can be divided into the following categories: multisensory signals, data augmentation, localglobal consistency, and temporal consistency.”
From Self-Supervised Learning for Time Series Analysis: Taxonomy, Progress, and Prospects (2024) · cites this paper as [29]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Contrastive learning [56-58] is in fact a training procedure used in recent artificial intelligence models as a way to efficiently learn semantic representations.”
From How Adversarial REM Dreams May Facilitate Creativity, and Why We Become Aware of Them (2024) · cites this paper as [56]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“As summarized in [29], commonly used contrastive loss functions generally include scoring functions (cosine similarity), energy-based margin functions (pair loss and triplet loss), probabilistic NCE-based functions, and mutual information based functions.”
From Self-Supervised Learning for Time Series Analysis: Taxonomy, Progress, and Prospects (2024) · cites this paper as [29]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 2×
“In SimCLRv2 [17], empirical experiments show that the output of the second layer of a 3-layer MLP projection head is a better representation for transfer learning than the output of the base ResNet [42] encoder.”
BackgroundMethodVaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017) · arXiv · doi:10.48550/arxiv.1706.03762 · Cited by 120,404 · cited 3×
“Recently, the Bidirectional Encoder Representation from Transformer (BERT) [24] model learns bidirectional word representations using the Transformer architecture’s decoder [103] and demonstrated great performance for transfer learning in multiple downstream tasks.”
MethodHochreiter, S., & Schmidhuber, J. (1997) · Neural Computation · doi:10.1162/neco.1997.9.8.1735 · Cited by 102,318 · cited 1×
“Even though the BERT [24] text encoder and the Faster RCNN [87] object detection model are trained separately and not updated by the contrastive loss, the LSTM cells [48] and a linear project head can still map between words and object representations.”
MethodExploiting Generative AI to Scale up Intelligent Tutoring Systems
Jakubův, J., Chvalovský, K., Goertzel, Z., Kaliszyk, C., Olšák, M., Piotrowski, B., Schulz, S., Suda, M., & Urban, J. (2023) · DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · doi:10.4230/lipics.itp.2023.19 · Cited by 76,658
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Ren, S., He, K., Girshick, R., & Sun, J. (2017) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2016.2577031 · Cited by 73,472
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805 · Cited by 73,173 · cited 5×
“The discretised speech representation can then be used directly as a representation for other models that expect discrete input such as BERT [24].”
MethodInfluentialFaster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Ren, S., He, K., Girshick, R., & Sun, J. (2015) · arXiv · doi:10.48550/arxiv.1506.01497 · Cited by 70,812
ImageNet: A large-scale hierarchical image database
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009) · 2009 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2009.5206848 · Cited by 65,333
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) · doi:10.18653/v1/n19-1423 · Cited by 54,325 · cited 5×
“The discretised speech representation can then be used directly as a representation for other models that expect discrete input such as BERT [24].”
MethodInfluentialEfficient Estimation of Word Representations in Vector Space
Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013) · arXiv · doi:10.48550/arxiv.1301.3781 · Cited by 26,568
Research works printed but not linked: 1
- stratos, 2019, Noise contrastive estimation
Cited by
Items 901–948 of 948
DFCE-KanT: predicting spatial gene expression from histology images via contrastive learning
Li, F., Wang, P., Shen, J., & Wu, Y. (2026) · Bioinformatics · doi:10.1093/bioinformatics/btag502
“Finally, layer normalization was applied to standardy feature distributions and eliminate batch effects, generating aligned features F x and F z for subsequent contrastive learning (Chen et al. 2020, Le-Khac et al. 2020).”
Jiang, C., & Ai, L. (2026) · Multimedia Systems · doi:10.1007/s00530-026-02497-2
“Furthermore, in the field of unsupervised image retrieval, the CL framework has attracted considerable attention due to its strong capability in exploiting unlabeled data for representation learning[24].”
Ismael, M. R., Abd, H. J., Al-Hamiri, M. G., Homod, R. Z., Mohammad, H. I., Sadeq, A. M., & Khatib, A. E. (2026) · IET Communications · doi:10.1049/cmu2.70191
Umamahesh, T. L. V., & Gampala, V. (2026) · International Research Journal of Multidisciplinary Scope · doi:10.47857/irjms.2026.v07i03.010171
“By maximizing agreement between variously augmented representations of the same sample while contrasting them against other samples, SimCLR and Momentum Contrast (MoCo) develop discriminative embeddings, making efficient use of vast pools of unlabeled data (17, 18).”
Zhang, A., Shi, J., Han, X., Gao, M., Wei, C., Wu, J., Zhang, M., Song, Y., & Liu, B. (2026) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2026.3713750
“Previous CL solutions can be categorized broadly into four groups: memory bank based [7], [8], [9], end-to-end embed-dings based [10], [11], [12], [13], [14], [15], cluster based [16], [17], [18], [19] and augmentation-driven approaches [20], [21], [22], [23].”
MT-ADA: active dual alignment for multi-target domain adaptation
Lin, Y., Deng, X., & Liang, X. (2026) · The Journal of Supercomputing · doi:10.1007/s11227-026-08727-8
“Furthermore, several approaches [12, 13, 18, 19, 35] integrate self-supervised and contrastive learning to exploit the latent structure in unlabeled target data, thereby boosting unsupervised domain adaptation performance.”
Time series anomaly detection models with contrastive representation learning
Quijada, F., Drobnyh, K., & Runger, G. C. (2026) · Journal of Intelligent Information Systems · doi:10.1007/s10844-026-01078-z
“The main components of a contrastive learning framework are: a stochastic data augmentation module, a base sequence encoder, a projection head, and the contrastive loss, which is used for learning (Chen et al., 2020b; Le-Khac et al., 2020).”
Representation Learning for Financial Time-Series Forecasting
Pawar, M. (2026) · International Journal of Adaptive Management and Business Intelligence · doi:10.67231/0x58ht41
“The learned representations are therefore more informative for downstream forecasting than raw prices or simple technical indicators, a pattern also observed when contrastive objectives are applied across other modalities and problem domains [12].”
Qi, J., Zhai, H., Wang, P., & Yang, F. (2026) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2026.115785
Guo, Q., Cao, L., Liua, X., Yang, K., Hou, C., Wang, B., Jia, J., Zhua, Z., & Duan, L. (2026) · Applied Computing and Geosciences · doi:10.1016/j.acags.2026.100391
A similarity-aware network with contrastive optimization for biomedical image segmentation
Lin, R., Yang, Z., Xu, Z., Gao, N., Yan, R., Shi, X., Lin, S., & Huang, B. (2026) · BMC Medical Imaging · doi:10.1186/s12880-026-02644-1
“Contrastive learning is a self-supervised learning (SSL) method that can learn image-level features from large-scale data without any manual labels [17, 18].”
The Geometry of Bias in Contrastive Embeddings: A Spectral Analysis
Ogbuokiri, B., Rahman, H., Martin, D., Sproxton, K., Belcastro, A., Huang, R., & Panchal, V. (2026) · Communications in computer and information science · doi:10.1007/978-3-032-22190-2_21
Dinh, H. M., Zong, W., Chow, Y.-W., & Susilo, W. (2026) · Applied Sciences · doi:10.3390/app16168323
Niyongabo, E., & Abdulrazak, B. (2026) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2026.111372
Knowledge distillation in medical image analysis: A survey
Chakrawarthige, O., Paluru, N., Padmanabhan, S., Aaqilah, A. J., Gowry Nandana, M., Gupta, U., Mathew, R. S., & Awasthi, N. (2026) · Array · doi:10.1016/j.array.2026.101167
Robust Sentiment Analysis with Minimal Preprocessing: An Adversarial-Contrastive Approach
Pedapapu, S. (2025) · NORMA
Jin, X., Wang, J., Ge, Y., Huang, X., & Chen, Y. (2026) · Network Modeling Analysis in Health Informatics and Bioinformatics · doi:10.1007/s13721-026-00867-4
Liang, T., Du, L., & Lan, G. (2026) · Ad Hoc Networks · doi:10.1016/j.adhoc.2026.104406
Multimodal Contrastive Learning for Zero-Shot HS Code Classification in Container Screening
Aljufri, M. A., Purwitasari, D., & Pradana, H. (2026) · IEEE Access · doi:10.1109/access.2026.3730511
“Unlike traditional discriminative models, contrastive frameworks align representations of the same concept across modalities by maximizing similarity between positive pairs and minimizing affinity between negative samples [15], [16].”
Wang, J., Luo, H., Zhang, G., Song, P., Wang, J., Wang, Y., Sun, Z., & Gao, S. (2026) · Pattern Recognition · doi:10.1016/j.patcog.2026.114861
Contrastive Learning for Aspect Representation Towards Explainable Recommendation
Hasan, E., & Ding, C. (2025) · IEEE/WIC/ACM International Conference on Web Intelligence and Intelligent Agent Technology (WI-IAT) · doi:10.1109/wi-iat67162.2025.00070
“To this end, we utilize the contrastive learning method [35] to learn representations of reviews at the aspect level and combine them with ID-based features to predict ratings and generate explanations.”
Hao, J., Zhang, S., Hou, C., & Xu, G. (2026) · Engineering Research Express · doi:10.1088/2631-8695/aea5e9
“Contrastive learning is an effective tool to solve the issue of feature learning with limited samples [20–22].”
Wang, W., Jiang, Z., Yang, S., Qin, Z., Shi, Y., Zhu, Y., & Sun, Z. (2026) · Foods · doi:10.3390/foods15183273
Contrastive learning with prototype reliability calibration for plant leaf recognition
Zeng, X., Cai, W., Lin, J., Yang, W., & Gao, Y. (2026) · Discover Artificial Intelligence · doi:10.1007/s44163-026-02196-x
“Contrastive learning provides a complementary path by shaping the geometry of the embedding space: positives are encouraged to cluster while negatives are pushed apart, leading to more discriminative features [30].”
Patient Aware Contrastive Learning for Paroxysmal Atrial Fibrillation Detection Using RR Intervals
Niroshana, Y., Wimalasiri, W. M., Kulasingham, J. P., & Hettiarachchi, C. (2026) · Moratuwa Engineering Research Conference (MERCon) · doi:10.1109/mercon71835.2026.11691431
Unsupervised clustering of tear damage in soybean kernels
Saurabh, P., Kanchwala, H., Miglani, A., & Kankar, P. K. (2026) · Journal of Food Measurement & Characterization · doi:10.1007/s11694-026-04912-8
“CRL [ 30] learns visual representations by maximising an agreement between differently augmented views of the same image.”
On the Value of Images to Learn Ecology-Aware Geographic Embeddings
Menoud, A., Servajean, M., Pavez, P. U., Larcher, T., & Marcos, D. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-39518-4_27
Graph conditional diffusion contrastive clustering for hyperspectral images
Wang, L., Liu, M., Xiao, L., Jiang, Y., Wang, H., & Hu, W. (2026) · Computer Vision and Image Understanding · doi:10.1016/j.cviu.2026.104974
Madgi, V. M., Gidnandi, S., Nisha, D., Kshitij, H., & Muttal, C. (2026) · Multimedia Tools and Applications · doi:10.1007/s11042-026-21961-9
“Originally formalized within self-supervised frameworks, the InfoNCE (Information Noise-Contrastive Estimation) objective [14, 15, 24 ] functions fundamentally as a multi-class categorical cross-entropy loss.”
Zhang, H., Wang, C., & Moon, J. (2026) · Machine Learning · doi:10.1007/s10994-026-07170-3
“This enables representations that not only maintain local smoothness but also form global semantic clusters (Le-Khac et al. 2020 ).”
CLLAP: Contrastive Learning-Based LiDAR-Augmented Pretraining for Enhanced Radar-Camera Fusion
Liu, B., Zhu, C., Xue, H., Teng, J., Liu, J., Wang, E., Dai, P., & Wang, P. (2026) · doi:10.1109/cvprf71909.2026.00014
“Self-supervised learning (SSL) leverages pseudo-labels to generate representations for downstream tasks, with contrastive learning emerging as a key technique in computer vision [17].”
Kohli, R., Bhabad, H. P., Patel, A. S., Rakhade, V. M., Patil, N. S., & Kadlag, V. (2026) · Information Dynamics and Applications · doi:10.56578/ida050204
Huang, X., Li, J., Zeng, Y., Huang, T., & Huang, H. (2026) · Quality and Reliability Engineering International · doi:10.1002/qre.70453
Park, S.-Y., & Kwak, K.-C. (2026) · Applied Sciences · doi:10.3390/app162010307
Rodgers, N. (2026) · Journal of The Royal Society Interface · doi:10.1098/rsif.2025.1243
Jiang, L., Zhang, Y., Li, Y., You, P., Zhang, X., Wang, L., & Guo, S. (2026) · Structural Health Monitoring · doi:10.1177/14759217261495871
Speech2Pro: Speech-driven in-situ projection authoring for stage performance
Hu, Y. '., Zhang, J., Wei, Z., Shi, Y., Hu, W., Cao, N., & Quigley, A. J. (2026) · International Journal of Human-Computer Studies · doi:10.1016/j.ijhcs.2026.103949
From Cloud to Edge: Deployable GeoAI for Cross-Hurricane Damage Assessment
Talreja, P. V., & Durbha, S. S. (2026) · IEEE International Geoscience and Remote Sensing Symposium (IGARSS) · doi:10.1109/igarss60577.2026.11730889
“There are many recent research studies [2, 3, 5, 6, 7] in deep learning and remote sensing done, which have enabled automated damage classification from post-event imagery.”
Numeric versus embedding pipelines for optimized clinical risk prediction
Ge, A., Wu, J., Wu, X., Balasubramanian, J. B., Nautiyal, V., Jayakumar, R., Murali, C., Lee, A., Nguyen, A., Allen, M., et al. (2026) · JAMIA Open · doi:10.1093/jamiaopen/ooag199
“28,29 General embedding objectives may favor broad clinical similarity over outcome-specific separation, but this mechanism cannot be tested in the present study, particularly because the pretraining corpora and objectives of several evaluated models are not publicly available.”
Representation Learning via Composite Subspace Flow MatchingPreprint
Afonso, J., & Figueiredo, M. A. T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.32256
“The primary objective of self-supervised representation learning is to extract informative, high-level, and useful features for downstream tasks from unlabeled raw data [1–3].”
Tai, L., Lin, X., Zhang, H., & Yuan, X. (2026) · Information · doi:10.3390/info17111070
Yao, X., Feng, Y., & Wang, Q. (2026) · Sensors · doi:10.3390/s26216861
Uendes, B., Gisbert, C. L., & Hoogendoorn, M. (2026) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2026.111731
Petković, H., Španjol, A., & Mihalić, D. (2023) · Data Engineering and Applications · doi:10.64972/dea.2023.v2i3.3715d:52-64
He, C., Sun, J., Zhang, Z., Sun, Y., & Wang, L. (2026) · IEEE International Conference on Signal Processing, Communications and Computing (ICSPCC) · doi:10.1109/icspcc70129.2026.11754352
A Study on Self-supervised Learning Objectives for Intrusion Detection
Rosa, D. D. L., del Jesús, M. J., & Charte, F. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-43170-7_14
Metadata
- Authors
- Le-Khac, P. H., Healy, G., & Smeaton, A. F.
- Affiliations
- Dublin City University (IE)
- Venue
- IEEE Access
- Published
- 2020
- Volume
- 8
- Pages
- 193907-193934
- ISSN
- 2169-3536 (ISSN-L)
- OpenAlex
- W3096655658
- Cited by
- 862
- References
- 144
- Citations by year
- 2026: 111 · 2025: 197 · 2024: 212 · 2023: 157 · 2022: 113 · 2021: 71 · 2020: 1
- FWCI
- 33.61
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Locations
- IEEE Access (publishedVersion, cc-by)
- arXiv (Cornell University) (acceptedVersion)
- Dublin City University Open Access Institutional Repository (Dublin City University) (submittedVersion, cc-by)
- Arrow@dit (Dublin Institute of Technology) (submittedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- arXiv (Cornell University) (cc-by)
- Funders
- Science Foundation Ireland through the Science Foundation Ireland (SFI) Centre for Research Training in Machine Learning (18/CRT/6183)
- Insight Centre for Data Analytics (SFI/12/RC/2289_P2)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning and Algorithms
- Keywords
- contrastive learning, representation learning, self-supervised learning, metric learning, inductive bias, computer vision, natural language processing, audio signal processing, reinforcement 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.
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.