The DOI matches its Crossref record; 34 of the 34 references it cites were found in a registry or an index; 2 of the 10 sources its claims cite were checked for support.
Time Series Change Point Detection with Self-Supervised Contrastive Predictive Coding
Open accessAbstract
Change Point Detection (CPD) methods identify the times associated with changes in the trends and properties of time series data in order to describe the underlying behaviour of the system. For instance, detecting the changes and anomalies associated with web service usage, application usage or human behaviour can provide valuable insights for downstream modelling tasks. We propose a novel approach for self-supervised Time Series Change Point detection method based on Contrastive Predictive coding (TS − CP2). TS − CP2 is the first approach to employ a contrastive learning strategy for CPD by learning an embedded representation that separates pairs of embeddings of time adjacent intervals from pairs of interval embeddings separated across time. Through extensive experiments on three diverse, widely used time series datasets, we demonstrate that our method outperforms five state-of-the-art CPD methods, which include unsupervised and semi-supervised approaches. TS − CP2 is shown to improve the performance of methods that use either handcrafted statistical or temporal features by 79.4% and deep learning-based methods by 17.0% with respect to the F1-score averaged across the three datasets.
Overview
CoChat Summary
TL;DR
The paper proposes TS-CP2, a self-supervised time series change point detection method that uses contrastive predictive coding to learn embeddings of adjacent time intervals. It reports improved F1-scores over five state-of-the-art CPD baselines across Yahoo! Benchmark, HASC, and USC-HAD.
Key findings
- TS-CP2 improved over methods using handcrafted statistical or temporal features by 79.4% and over deep learning-based methods by 17.0% in F1-score averaged across the three datasets.
- On HASC, TS-CP2 achieved the highest average F1-score with improvements of 19.2%, 54.8%, 10.1%, 11.1%, and 3.8% over FLOSS, aHSIC, RuLSIF, ESPRESSO, and KL-CPD, respectively.
- Best reported F1-scores were 0.8428 on Yahoo! Benchmark at a 75-sample detection margin, 0.8571 on USC-HAD at a 200-sample margin, and 0.6316 on HASC at a 200-sample margin.
- The paper states TS-CP2 is the first CPD approach based on contrastive representation learning and the first to use contrastive learning on general time series.
Methods
- Design: self-supervised contrastive predictive coding with a WaveNet/TCN encoder (two TCN stacks, 64 filters, kernel size 4, dilations 1, 4, 16) followed by a three-layer projection head; positive pairs are contiguous history/future windows, negatives are windows separated across time; training uses InfoNCE/N-paired loss.
- Data: three datasets—Yahoo! Benchmark (100 sequences, 164K samples, 208 change points), HASC (39K samples, 65 change points), and USC-HAD (97K samples, 30 change points).
- Analysis: change points are estimated by cosine similarity between history and future window embeddings, then peak finding on the difference between similarity and its moving average; performance is evaluated by F1-score across specified detection margins and compared against ESPRESSO, FLOSS, aHSIC, RuLSIF, and KL-CPD.
Limitations
- Stated: HASC was the most challenging dataset because its high change-point density (65 change points in about 39K samples) increases the chance that positive sample pairs contain change points; since training is self-supervised, ground truth cannot be used to correct these false positive pairs, and the authors suggest light negative mining or augmentation.
- Stated: the minimum temporal distance threshold for negative sampling must be set based on the time series application.
- Evident from the text: experiments use only accelerometer data for HASC and USC-HAD, and only Benchmark 4 of Yahoo! is used, so generalization to other sensor channels or benchmark subsets is not tested.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
2 Valid · 2 Not valid · 7 Not checked · 49 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“Current CPD methods have failed to generalise effectively [13] as the semantic boundaries of different applications will usually be associated with different time series properties.”
What the sources say
- Supportedcites [13] source
The Introduction (p. 2) states that pervasive-computing time-series are heterogeneous and that 'the semantics of each use case should be extracted by exploiting different time-series properties', which is exactly the reason clause the citing sentence gives. The Related Work/Introduction passages carry the corresponding critique of existing change-point/segmentation methods: parametric statistical approaches 'have somewhat limited application given they impose strong assumptions upon the statistical properties of the time-series' (p. 3), current approaches suffer from 'Model assumptions ... that can limit its application' (p. 3), shape-based assumptions 'can lead to degraded segmentation performance' (p. 3), and non-parametric approaches 'are not universally applicable to HAR applications' (p. 6). Together our copy clearly treats current CPD/segmentation methods as failing to transfer across applications, though it phrases this as limited/universal applicability in wearable-sensing contexts rather than using the word 'generalise'. (LLM, conf=0.80)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“𝑇𝑆 -𝐶𝑃 2 is shown to improve the performance of methods that use either handcrafted statistical or temporal features by 79.4% and deep learning-based methods by 17.0% with respect to the F1-score averaged across the three datasets.”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Through extensive experiments on three diverse, widely used time series datasets, we demonstrate that our method outperforms five state-of-the-art CPD methods, which include unsupervised and semi-supervised approaches.”
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“𝑇𝑆 -𝐶𝑃 2 is shown to improve the performance of methods that use either handcrafted statistical or temporal features by 79.4% and deep learning-based methods by 17.0% with respect to the F1-score averaged across the three datasets.”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“It is clear that the local minima of the difference between the cosine similarity of each interval pair ( ℎ 𝑖 , 𝑓 𝑖 ) and the average similarity across recent intervals pairs coincide with true change points.”
boostedstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
33 more claims
Create a free account to see all 60 claims and to verify the 8 references still open.
- Probably not a substantial claim. 13
- It may state two separate claims as one. 11
- The quote only partly states this claim. 2
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 cited source says what the claim attributes to it.
In the citing paper
“Following the architecture of CPC, LNT [110], TRL-CPC [111], TS-CP 2 [112], and Skip-Step CPC [113] were proposed.”
From Self-Supervised Learning for Time Series Analysis: Taxonomy, Progress, and Prospects (2024) · cites this paper as [112]
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.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 3.4 Change Point Detection Module · p. 6
“The time intervals associated with these local maxima are considered as the change point estimates.”
§ 3.4 Change Point Detection Module · p. 6
“It is clear that the local minima of the difference between the cosine similarity of each interval pair ( ℎ 𝑖 , 𝑓 𝑖 ) and the average similarity across recent intervals pairs coincide with true change points.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 4.5 Baseline Comparison · p. 9
“The HASC dataset is the smallest in size (around 39K samples) but contains the largest number of change points (65 change points in total).”
§ 4.1 Datasets · p. 7
“| Yahoo! Benchmark | 164K | 100 | 1 | 208 | | HASC | 39K | 1 | 3 | 65 |”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 5 CONCLUSION · p. 10
“We evaluated the ability of 𝑇𝑆 -𝐶𝑃 2 in detecting change points against six other well-known state-of-the-art methods across three datasets.”
§ 4.2 Baseline Methods · p. 7
“The performance of the proposed 𝑇𝑆 -𝐶𝑃 2 method was compared against five state-of-the-art unsupervised change point detection techniques that included ESPRESSO [13], FLOSS [18], aHSIC [52], RuLSIF [30], and KL-CPD [6].”
Citations
Cites
FaceNet: A unified embedding for face recognition and clustering
Schroff, F., Kalenichenko, D., & Philbin, J. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298682 · Cited by 15,267 · cited 1×
“A number of sampling strategies have been proposed, including hard negative sampling [43], semi-hard mining [41], distance weighted”
BackgroundMulti-scale context aggregation by dilated convolutions
Yu, F., & Koltun, V. (2015) · arXiv · doi:10.48550/arxiv.1511.07122 · Cited by 7,738 · cited 1×
“Besides this, exponentially dilated convolutions have also been introduced to better capture, compared to full convolutions, long-range dependencies at constant depth by exponentially increasing the receptive field of the network [4, 32, 54].”
BackgroundLearning a Similarity Metric Discriminatively, with Application to Face Verification
Chopra, S., Hadsell, R., & LeCun, Y. (2005) · 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05) · doi:10.1109/cvpr.2005.202 · Cited by 4,036 · cited 1×
“s considered to be different). A representation is learnt to bring the positive sample pairs closer together and to further separate negative sample pairs within the embedding space. Contrastive loss [9] and Triplet loss [48] are the most commonly used loss functions. In general, the triplet loss function outperforms…”
BackgroundData-Efficient Image Recognition with Contrastive Predictive Coding
Hénaff, O. J., Srinivas, A., De Fauw, J., Razavi, A., Doersch, C., Eslami, S. M. A., & Oord, A. V. D. (2019) · arXiv · doi:10.48550/arxiv.1905.09272 · Cited by 1,014 · cited 6×
“encode the time series, given it has been shown that temporal convolutional networks (TCN) can often produce superior prediction performance with sequential data [21] and are generally easier to train.”
BackgroundMethodInfluentialSampling Matters in Deep Embedding Learning
Wu, C.-Y., Manmatha, R., Smola, A. J., & Krahenbuhl, P. (2017) · 2017 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.309 · Cited by 836 · cited 1×
“nces for training. This issue of complexity has been addressed by Hard Negative Instance Mining, which has been shown to play a critical role in ensuring contrastive cost functions are more efficient [15, 49]. A number of sampling strategies have been proposed, including hard negative sampling [43], semi-hard mining […”
Discriminative Learning of Deep Convolutional Feature Point Descriptors
Simo-Serra, E., Trulls, E., Ferraz, L., Kokkinos, I., Fua, P., & Moreno-Noguer, F. (2015) · 2015 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2015.22 · Cited by 804 · cited 1×
“to play a critical role in ensuring these cost functions are effective [14, 47]. A number of different sampling strategies have been proposed to addresses this issue, including hard negative sampling [40], semihard mining [38], distance weighted sampling [47], hard negative class mining [41], and rank-based negative m…”
BackgroundInfluentialThe Opportunity challenge: A benchmark database for on-body sensor-based activity recognition
Chavarriaga, R., Sagha, H., Calatroni, A., Digumarti, S. T., Tröster, G., Millán, J. D. R., & Roggen, D. (2013) · Pattern Recognition Letters · doi:10.1016/j.patrec.2012.12.014 · Cited by 742 · cited 1×
“analysis [27], IoT applications and smart homes [3], human activity recognition (HAR) [5, 13, 29, 42, 46], human physiological and emotional analysis [13], factory automation [51], trajectory prediction [37], user authentication [22], life-logging [7], elderly rehabilitation [28], and daily work routine studies [12].”
BackgroundMS-TCN: Multi-Stage Temporal Convolutional Network for Action Segmentation
Farha, Y. A., & Gall, J. (2019) · 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2019.00369 · Cited by 673 · cited 3×
“Kernel Size = 4, Dilation = [1, 4, 16] Layout of each block of TCN Prediction/projection head”
BackgroundMethodUSC-HAD: a daily activity dataset for ubiquitous activity recognition using wearable sensors
Zhang, M., & Sawchuk, A. A. (2012) · Proceedings of the 2012 ACM Conference on Ubiquitous Computing · doi:10.1145/2370216.2370438 · Cited by 554 · cited 1×
“• USC-HAD 5 [56].”
Change-point detection in time-series data by relative density-ratio estimation
Liu, S., Yamada, M., Collier, N., & Sugiyama, M. (2013) · Neural Networks · doi:10.1016/j.neunet.2013.01.012 · Cited by 447 · cited 7×
“The methods in RuLSIF [30], KLIEP [54] and SEP [3] used a non-parametric Gaussian kernel to model the density ratio distribution between subsequent time intervals.”
Influential
Research works printed but not linked: 23
- Samaneh Aminikhanghahi and Diane J. Cook . 2017. A Survey of Methods for Time Series Change Point Detection. Knowledge and information systems 51, 2 (01 May 2017 ), 339–367. Samaneh Aminikhanghahi and Diane J. Cook. 2017. A Survey of Methods for Time Series Change Point Detection. Knowledge and information systems 51, 2 (01 May 2017), 339–367.
- Samaneh Aminikhanghahi and Diane J Cook . 2019. Enhancing Activity Recognition Using CPD-based Activity Segmentation. Pervasive and Mobile Computing 53 ( 2019 ). Samaneh Aminikhanghahi and Diane J Cook. 2019. Enhancing Activity Recognition Using CPD-based Activity Segmentation. Pervasive and Mobile Computing 53 (2019).
- Michelle Basseville and Igor V Nikiforov . 1993. Detection of abrupt changes: theory and application . Prentice Hall . Michelle Basseville and Igor V Nikiforov. 1993. Detection of abrupt changes: theory and application. Prentice Hall.
- Wei-Cheng Chang Chun-Liang Li Yiming Yang and Barnabás Póczos. 2019. Kernel change-point detection with auxiliary deep generative models. arXiv preprint arXiv:1901.06077(2019). Wei-Cheng Chang Chun-Liang Li Yiming Yang and Barnabás Póczos. 2019. Kernel change-point detection with auxiliary deep generative models. arXiv preprint arXiv:1901.06077(2019). · arXiv
- Ting Chen , Simon Kornblith , Mohammad Norouzi , and Geoffrey Hinton . 2020. A Simple Framework for Contrastive Learning of Visual Representations. ICML ( 2020 ). Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A Simple Framework for Contrastive Learning of Visual Representations. ICML (2020).
- Tim De Ryck Maarten De Vos and Alexander Bertrand. 2020. Change Point Detection in Time Series Data using Autoencoders with a Time-Invariant Representation. arXiv preprint arXiv:2008.09524(2020). Tim De Ryck Maarten De Vos and Alexander Bertrand. 2020. Change Point Detection in Time Series Data using Autoencoders with a Time-Invariant Representation. arXiv preprint arXiv:2008.09524(2020). · arXiv
- Shohreh Deldari , Jonathan Liono , Flora D Salim , and Daniel V Smith . 2019 . Inferring Work Routines and Behavior Deviations with Life-logging Sensor Data . In Proceedings of ACM International Conference on Web Search and Data Mining (WSDM) workshop on Task Intelligence (TI@WSDM) (2019). ACM. Shohreh Deldari, Jonathan Liono, Flora D Salim, and Daniel V Smith. 2019. Inferring Work Routines and Behavior Deviations with Life-logging Sensor Data. In Proceedings of ACM International Conference on Web Search and Data Mining (WSDM) workshop on Task Intelligence (TI@WSDM) (2019). ACM.
- Jean-Yves Franceschi Aymeric Dieuleveut and Martin Jaggi. 2019. Unsupervised scalable representation learning for multivariate time series. In Advances in Neural Information Processing Systems. 4650–4661. Jean-Yves Franceschi Aymeric Dieuleveut and Martin Jaggi. 2019. Unsupervised scalable representation learning for multivariate time series. In Advances in Neural Information Processing Systems. 4650–4661.
- Michael Gutmann and Aapo Hyvärinen . 2010 . Noise-contrastive estimation: A new estimation principle for unnormalized statistical models . In Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics. 297–304 . Michael Gutmann and Aapo Hyvärinen. 2010. Noise-contrastive estimation: A new estimation principle for unnormalized statistical models. In Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics. 297–304.
- David Tse Jung Huang , Yun Sing Koh , Gillian Dobbie , and Russel Pears . 2014 . Detecting Changes in Rare Patterns from Data Streams . In Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD). Springer, 437–448 . David Tse Jung Huang, Yun Sing Koh, Gillian Dobbie, and Russel Pears. 2014. Detecting Changes in Rare Patterns from Data Streams. In Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD). Springer, 437–448.
- Tero Kokkonen , Samir Puuska , Janne Alatalo , Eppu Heilimo , and Antti Mäkelä . 2019. Network anomaly detection based on wavenet . In Internet of Things, Smart Spaces, and Next Generation Networks and Systems . Springer , 424–433. Tero Kokkonen, Samir Puuska, Janne Alatalo, Eppu Heilimo, and Antti Mäkelä. 2019. Network anomaly detection based on wavenet. In Internet of Things, Smart Spaces, and Next Generation Networks and Systems. Springer, 424–433.
- Agnes WK Lam , Danniel Varona-Marin , Yeti Li , Mitchell Fergenbaum , and Dana Kulić . 2016. Automated Rehabilitation System: Movement Measurement and Feedback for Patients and Physiotherapists in the Rehabilitation Clinic. Human–Computer Interaction 31, 3-4 ( 2016 ), 294–334. Agnes WK Lam, Danniel Varona-Marin, Yeti Li, Mitchell Fergenbaum, and Dana Kulić. 2016. Automated Rehabilitation System: Movement Measurement and Feedback for Patients and Physiotherapists in the Rehabilitation Clinic. Human–Computer Interaction 31, 3-4 (2016), 294–334.
- Andriy Mnih and Koray Kavukcuoglu. 2013. Learning word embeddings efficiently with noise-contrastive estimation. In Advances in neural information processing systems. 2265–2273. Andriy Mnih and Koray Kavukcuoglu. 2013. Learning word embeddings efficiently with noise-contrastive estimation. In Advances in neural information processing systems. 2265–2273.
- Ramy Mounir Roman Gula Jörn Theuerkauf and Sudeep Sarkar. 2020. Temporal Event Segmentation using Attention-based Perceptual Prediction Model for Continual Learning. arXiv preprint arXiv:2005.02463(2020). Ramy Mounir Roman Gula Jörn Theuerkauf and Sudeep Sarkar. 2020. Temporal Event Segmentation using Attention-based Perceptual Prediction Model for Continual Learning. arXiv preprint arXiv:2005.02463(2020). · arXiv
- Aaron van den Oord , Sander Dieleman , Heiga Zen , Karen Simonyan , Oriol Vinyals , Alex Graves , Nal Kalchbrenner , Andrew Senior , and Koray Kavukcuoglu . 2016 . Wavenet: A generative model for raw audio. arXiv preprint arXiv:1609.03499(2016). Aaron van den Oord, Sander Dieleman, Heiga Zen, Karen Simonyan, Oriol Vinyals, Alex Graves, Nal Kalchbrenner, Andrew Senior, and Koray Kavukcuoglu. 2016. Wavenet: A generative model for raw audio. arXiv preprint arXiv:1609.03499(2016). · arXiv
- Aaron van den Oord Yazhe Li and Oriol Vinyals. 2018. Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748(2018). Aaron van den Oord Yazhe Li and Oriol Vinyals. 2018. Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748(2018). · arXiv
- Aaqib Saeed David Grangier and Neil Zeghidour. 2020. Contrastive Learning of General-Purpose Audio Representations. arXiv preprint arXiv:2010.10915(2020). Aaqib Saeed David Grangier and Neil Zeghidour. 2020. Contrastive Learning of General-Purpose Audio Representations. arXiv preprint arXiv:2010.10915(2020). · arXiv
- A. Saeed F. D. Salim T. Ozcelebi and J. Lukkien. 2020. Federated Self-Supervised Learning of Multi-Sensor Representations for Embedded Intelligence. IEEE Internet of Things Journal(2020) 1–1. A. Saeed F. D. Salim T. Ozcelebi and J. Lukkien. 2020. Federated Self-Supervised Learning of Multi-Sensor Representations for Embedded Intelligence. IEEE Internet of Things Journal(2020) 1–1.
- Kihyuk Sohn. 2016. Improved deep metric learning with multi-class n-pair loss objective. In Advances in neural information processing systems. 1857–1865. Kihyuk Sohn. 2016. Improved deep metric learning with multi-class n-pair loss objective. In Advances in neural information processing systems. 1857–1865.
- Zijun Wei Boyu Wang Minh Hoai Nguyen Jianming Zhang Zhe Lin Xiaohui Shen Radomír Mech and Dimitris Samaras. 2018. Sequence-to-segment networks for segment detection. In Advances in Neural Information Processing Systems. 3507–3516. Zijun Wei Boyu Wang Minh Hoai Nguyen Jianming Zhang Zhe Lin Xiaohui Shen Radomír Mech and Dimitris Samaras. 2018. Sequence-to-segment networks for segment detection. In Advances in Neural Information Processing Systems. 3507–3516.
- Kilian Q Weinberger and Lawrence K Saul . 2009. Distance metric learning for large margin nearest neighbor classification.Journal of Machine Learning Research 10, 2 ( 2009 ). Kilian Q Weinberger and Lawrence K Saul. 2009. Distance metric learning for large margin nearest neighbor classification.Journal of Machine Learning Research 10, 2 (2009).
- Renjie Wu and Eamonn J Keogh. 2020. Current Time Series Anomaly Detection Benchmarks are Flawed and are Creating the Illusion of Progress. arXiv preprint arXiv:2009.13807(2020). Renjie Wu and Eamonn J Keogh. 2020. Current Time Series Anomaly Detection Benchmarks are Flawed and are Creating the Illusion of Progress. arXiv preprint arXiv:2009.13807(2020). · arXiv
- Makoto Yamada , Akisato Kimura , Futoshi Naya , and Hiroshi Sawada . 2013 . Change-point Detection with Feature Selection in High-dimensional Time-series Data . In Proc. of 23th International Joint Conference on Artificial Intelligence (IJCAI). Makoto Yamada, Akisato Kimura, Futoshi Naya, and Hiroshi Sawada. 2013. Change-point Detection with Feature Selection in High-dimensional Time-series Data. In Proc. of 23th International Joint Conference on Artificial Intelligence (IJCAI).
Cited by
DCdetector: Dual Attention Contrastive Representation Learning for Time Series Anomaly Detection
Yang, Y., Zhang, C., Zhou, T., Wen, Q., & Sun, L. (2023) · Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining · doi:10.1145/3580305.3599295 · Cited by 387
“We compare our model with the 26 baselines for comprehensive evaluations, including the reconstruction-based model: AutoEncoder [58], LSTM-VAE [50], OmniAnomaly [63], BeatGAN [87], InterFusion [45], Anomaly Transformer [71]; the autoregressionbased models: VAR [4], Autoregression [55], LSTM-RNN [8], LSTM [32], CL-MPPC…”
MethodSelf-Supervised Learning for Time Series Analysis: Taxonomy, Progress, and Prospects
Zhang, K., Wen, Q., Zhang, C., Cai, R., Jin, M., Liu, Y., Zhang, J. Y., Liang, Y., Pang, G., Song, D., et al. (2024) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2024.3387317 · Cited by 295
“Following the architectureofCPC,LNT[110],TRL-CPC[111],TS-CP 2 [112], and Skip-Step CPC [113] were proposed.”
MethodCalibrated One-Class Classification for Unsupervised Time Series Anomaly Detection
Xu, H., Wang, Y., Jian, S., Liao, Q., Wang, Y., & Pang, G. (2024) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2024.3393996 · Cited by 162
“They assign class labels to different augmentation operations (e.g., adding noise, reversing, scaling, and smoothing) [27], neural transformations [36], contiguous and separate time segments [19], or different time resolutions [28].”
BackgroundMethodA Multihead Attention Self-Supervised Representation Model for Industrial Sensors Anomaly Detection
Qiao, Y., Lü, J., Wang, T., Liu, K., Zhang, B., & Snoussi, H. (2023) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2023.3280337 · Cited by 132
ColloSSL: Collaborative Self-Supervised Learning for Human Activity Recognition
Jain, Y., Tang, C. I., Min, C., Kawsar, F., & Mathur, A. (2022) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3517246 · Cited by 113
“SSL has also been used in the context of change point detection [11] and federated learning [47].”
Method
Metadata
- Authors
- Deldari, S., Smith, D. V., Xue, H., & Salim, F. D.
- Affiliations
- RMIT University (AU)
- Commonwealth Scientific and Industrial Research Organisation (AU)
- ORCID
- Deldari, Shohreh
- Venue
- Proceedings of the Web Conference 2021
- Publisher
- ACM
- Published
- 2021-04-19
- Pages
- 3124-3135
- OpenAlex
- W3109518304
- Cited by
- 136
- References
- 31
- Citations by year
- 2026: 11 · 2025: 33 · 2024: 46 · 2023: 24 · 2022: 20 · 2021: 2
- FWCI
- 14.00
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- International World Wide Web Conference (WWW) (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Signal Processing › Time Series Analysis and Forecasting
- Physical Sciences › Computer Science › Artificial Intelligence › Anomaly Detection Techniques and Applications
- Social Sciences › Decision Sciences › Statistics, Probability and Uncertainty › Advanced Statistical Process Monitoring
- Keywords
- contrastive predictive coding, self-supervised learning, contrastive learning, time series anomaly detection
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).
Cited sources are checked with a language model’s help (qwen3.8-omni-flash).
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Time Series Change Point Detection with Self-Supervised Contrastive Predictive Coding. https://app.cochat.ai/papers/10.1145/3442381.3449903
- The paper
- Deldari, S., Smith, D. V., Xue, H., & Salim, F. D. (2021). Time Series Change Point Detection with Self-Supervised Contrastive Predictive Coding. Proceedings of the Web Conference 2021. https://doi.org/10.1145/3442381.3449903
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.