pith. sign in

arxiv: 2210.09817 · v3 · submitted 2022-10-18 · 💻 cs.LG

Universal hidden monotonic trend estimation with contrastive learning

Edouard Pineau , S\'ebastien Razakarivony This is my paper

Pith reviewed 2026-05-24 10:56 UTC · model grok-4.3

classification 💻 cs.LG
keywords monotonic trend estimationcontrastive learningtime seriestrend detectionMann-Kendall testuniversal methodtemporal data
0
0 comments X

The pith

A contrastive learning method called CTE extracts any hidden monotonic trend from temporal data of any type without standard statistical assumptions.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents contrastive trend estimation as a way to pull out monotonic trends hidden in sequences of vectors, images, graphs or time series. It builds on ideas from the Mann-Kendall test but drops the usual requirements about data distribution or independence. Because the approach works on raw inputs of many kinds, it opens trend detection to data sets that break conventional tests. A reader would care if the method truly separates the trend factor in settings where older techniques give no answer or give wrong answers.

Core claim

We propose contrastive trend estimation (CTE), a method related to the Mann-Kendall test, that identifies any hidden trend underlying temporal data while avoiding the standard assumptions used for monotonic trend identification. In particular, CTE can take any type of temporal data (vector, images, graphs, time series, etc.) as input.

What carries the argument

Contrastive trend estimation (CTE), a contrastive learning setup that isolates the monotonic trend factor from the input data.

If this is right

  • Monotonic trend detection extends directly to image sequences and graph sequences.
  • No requirement for normality, independence, or other distributional assumptions remains.
  • The same procedure applies unchanged to vector-valued time series and standard scalar series.
  • Trend estimation becomes feasible on data modalities where Mann-Kendall tests cannot even be formulated.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The approach may allow trend analysis on multimodal streams that mix images and sensor readings.
  • If the contrastive pairs are chosen poorly the extracted factor could capture something other than monotonicity.
  • The method supplies a route to test for monotonicity in domains such as video or network traffic where classical tests are rarely used.

Load-bearing premise

A contrastive learning setup can be built that isolates the monotonic trend factor from arbitrary temporal inputs without relying on the usual statistical assumptions.

What would settle it

Run CTE on a collection of image sequences that contain a known monotonic change in pixel intensity; if the method returns no trend or returns a trend that does not match the known change, the claim fails.

Figures

Figures reproduced from arXiv: 2210.09817 by Edouard Pineau, S\'ebastien Razakarivony.

Figure 1
Figure 1. Figure 1: Illustration of the context of the paper’s contribution. We have a monitored system S that generates data samples (colored curves) at random time. The hidden trend τ underlying the system (colors from green to red) represents the hidden state of S that changes monotonically until a state restoration is applied (tools in hexagons): samples between two state restorations form a sequence with a monotonic hidd… view at source ↗
Figure 2
Figure 2. Figure 2: In survival analysis, several systems enter a study by giving a data sample. At [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Example of CTE on CMAPSS data with noisy trend. [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
read the original abstract

In this paper, we describe a universal method for extracting the underlying monotonic trend factor from time series data. We propose an approach related to the Mann-Kendall test, a standard monotonic trend detection method and call it contrastive trend estimation (CTE). We show that the CTE method identifies any hidden trend underlying temporal data while avoiding the standard assumptions used for monotonic trend identification. In particular, CTE can take any type of temporal data (vector, images, graphs, time series, etc.) as input. We finally illustrate the interest of our CTE method through several experiments on different types of data and problems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The manuscript introduces Contrastive Trend Estimation (CTE), a contrastive-learning approach inspired by the Mann-Kendall test, that claims to extract any hidden monotonic trend from arbitrary temporal inputs (vectors, images, graphs, time series) while avoiding the statistical assumptions of conventional monotonic-trend methods.

Significance. If the universality claim were rigorously established, the method would supply a modality-agnostic tool for trend extraction that could be applied directly to raw high-dimensional data in machine-learning pipelines. The contrastive formulation itself is a novel angle on a classical statistical task.

major comments (1)
  1. [Abstract] Abstract: the central assertion that CTE 'identifies any hidden trend underlying temporal data' for 'any type of temporal data (vector, images, graphs, time series, etc.)' while 'avoiding the standard assumptions' is not accompanied by a theorem, derivation, or formal argument showing that the contrastive positive/negative pair construction isolates the monotonic factor independently of modality-specific embedding or sampling choices.
minor comments (1)
  1. The abstract states that the method is illustrated 'through several experiments on different types of data and problems' yet supplies neither experimental protocol, quantitative metrics, nor baseline comparisons, preventing assessment of whether the claimed universality is realized in practice.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the major comment on the abstract below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central assertion that CTE 'identifies any hidden trend underlying temporal data' for 'any type of temporal data (vector, images, graphs, time series, etc.)' while 'avoiding the standard assumptions' is not accompanied by a theorem, derivation, or formal argument showing that the contrastive positive/negative pair construction isolates the monotonic factor independently of modality-specific embedding or sampling choices.

    Authors: The CTE construction defines positive pairs as data instances sharing the same underlying monotonic trend value and negative pairs as those with differing trend values. This definition is deliberately modality-agnostic and mirrors the rank-based, non-parametric logic of the Mann-Kendall test; the contrastive loss then trains an encoder to separate the trend factor from other sources of variation. Because the pair labels depend only on the (latent) trend coordinate and not on the input representation, the isolation step itself does not embed modality-specific assumptions. We acknowledge that the manuscript presents this argument motivationally rather than via a formal theorem. We will revise the abstract to moderate the universality phrasing and add a short subsection in the methods that explicitly derives the pair-construction step and discusses its independence from embedding architecture and sampling details. revision: partial

Circularity Check

0 steps flagged

No circularity: derivation does not reduce to self-definition or fitted inputs

full rationale

The provided abstract and context describe CTE as a contrastive-learning method inspired by but distinct from the Mann-Kendall test, with a claim of universality across modalities. No equations, parameter-fitting steps, or self-citations are exhibited that would make any 'prediction' or 'result' equivalent to its inputs by construction. The central claim rests on an empirical transfer from contrastive objectives to trend isolation, which may be unproven but is not shown to be tautological. No load-bearing step matches any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no specific free parameters, axioms, or invented entities can be identified from the provided information.

pith-pipeline@v0.9.0 · 5623 in / 1021 out tokens · 59334 ms · 2026-05-24T10:56:46.891152+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

49 extracted references · 49 canonical work pages · 2 internal anchors

  1. [1]

    Development of a new method of wavelet aided trend detection and estimation

    Kaz Adamowski, Andreas Prokoph, and Jan Adamowski. Development of a new method of wavelet aided trend detection and estimation. Hydrological Processes: An Interna- tional Journal, 23(18):2686–2696, 2009

    work page 2009

  2. [2]

    and Poveda Germ´ an

    Carmona Alejandra M. and Poveda Germ´ an. Detection of long-term trends in monthly hydro-climatic series of colombia through empirical mode decomposition. Climatic Change, 123:301–313, 2014

    work page 2014

  3. [3]

    Self-supervised representation learning from electroencephalography signals

    Hubert Banville, Graeme Moffat, Isabela Albuquerque, Denis-Alexander Engemann, Aapo Hyv¨ arinen, and Alexandre Gramfort. Self-supervised representation learning from electroencephalography signals. In 2019 IEEE 29th International Workshop on Machine Learning for Signal Processing (MLSP) , pages 1–6. IEEE, 2019

    work page 2019

  4. [4]

    Representation learning: A review and new perspectives

    Yoshua Bengio, Aaron Courville, and Pascal Vincent. Representation learning: A review and new perspectives. IEEE transactions on pattern analysis and machine intelligence, 35(8):1798–1828, 2013

    work page 2013

  5. [5]

    Analysis of survival data by the proportional odds model

    Steve Bennett. Analysis of survival data by the proportional odds model. Statistics in medicine, 2(2):273–277, 1983

    work page 1983

  6. [6]

    Independent slow feature analysis and nonlinear blind source separation

    Tobias Blaschke, Tiziano Zito, and Laurenz Wiskott. Independent slow feature analysis and nonlinear blind source separation. Neural computation, 19(4):994–1021, 2007

    work page 2007

  7. [7]

    Design of multivariate alarm systems based on online calculation of variational directions

    Kuang Chen and Jiandong Wang. Design of multivariate alarm systems based on online calculation of variational directions. Chemical Engineering Research and Design, 122:11–21, 2017

    work page 2017

  8. [8]

    Blind source separation and independent component analysis: A review

    Seungjin Choi, Andrzej Cichocki, Hyung-Min Park, and Soo-Young Lee. Blind source separation and independent component analysis: A review. Neural Information Processing-Letters and Reviews, 6(1):1–57, 2005. 10

    work page 2005

  9. [9]

    Regression models and life-tables

    David R Cox. Regression models and life-tables. Journal of the Royal Statistical Society: Series B (Methodological), 34(2):187–202, 1972

    work page 1972

  10. [10]

    PySurvival: Open source package for survival analysis modeling, 2019–

    Stephane Fotso et al. PySurvival: Open source package for survival analysis modeling, 2019–

    work page 2019

  11. [11]

    Unsupervised scalable representation learning for multivariate time series

    Jean-Yves Franceschi, Aymeric Dieuleveut, and Martin Jaggi. Unsupervised scalable representation learning for multivariate time series. In Advances in Neural Information Processing Systems, pages 4650–4661, 2019

    work page 2019

  12. [12]

    Robust parameter estimation with a small bias against heavy contamination

    Hironori Fujisawa and Shinto Eguchi. Robust parameter estimation with a small bias against heavy contamination. Journal of Multivariate Analysis , 99(9):2053–2081, 2008

    work page 2053

  13. [13]

    Robust Loss Functions under Label Noise for Deep Neural Networks

    Aritra Ghosh, Himanshu Kumar, and PS Sastry. Robust loss functions under label noise for deep neural networks. arXiv preprint arXiv:1712.09482 , 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  14. [14]

    Richard O. Gilbert. Statistical Methods for Environmental Pollution Monitoring. 1987

    work page 1987

  15. [15]

    Econ 2 0a: Sufficiency, minimal sufficiency and the exponential family of distributions

    Chuan Goh. Econ 2 0a: Sufficiency, minimal sufficiency and the exponential family of distributions. 2001

    work page 2001

  16. [16]

    Monitoring for conservation and ecology , volume 3

    Frank Barrie Goldsmith. Monitoring for conservation and ecology , volume 3. Springer Science & Business Media, 2012

    work page 2012

  17. [17]

    Comparison of trend detection methods

    Katharine Lynn Gray. Comparison of trend detection methods. 2007

    work page 2007

  18. [18]

    Likelihood- free inference via classification

    Michael U Gutmann, Ritabrata Dutta, Samuel Kaski, and Jukka Corander. Likelihood- free inference via classification. Statistics and Computing , 28(2):411–425, 2018

    work page 2018

  19. [19]

    A survey of label-noise representation learning: Past, present and future

    Bo Han, Quanming Yao, Tongliang Liu, Gang Niu, Ivor W Tsang, James T Kwok, and Masashi Sugiyama. A survey of label-noise representation learning: Past, present and future. arXiv preprint arXiv:2011.04406 , 2020

    work page arXiv 2011

  20. [20]

    Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors

    Frank E Harrell Jr, Kerry L Lee, and Daniel B Mark. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Statistics in medicine , 15(4):361–387, 1996

    work page 1996

  21. [21]

    10 structural time series models

    Andrew C Harvey and Neil Shephard. 10 structural time series models. 1993

    work page 1993

  22. [22]

    The empirical mode decomposition and the hilbert spectrum for nonlinear and non-stationary time series analysis

    Norden E Huang, Zheng Shen, Steven R Long, Manli C Wu, Hsing H Shih, Quanan Zheng, Nai-Chyuan Yen, Chi Chao Tung, and Henry H Liu. The empirical mode decomposition and the hilbert spectrum for nonlinear and non-stationary time series analysis. Proceedings of the Royal Society of London. Series A: mathematical, physical and engineering sciences, 454(1971):...

    work page 1971

  23. [23]

    Meaningful trend in climate time series: A discussion based on linear and smoothing techniques for drought analysis in taiwan

    Shih-Han Huang, Khalid Mahmud, and Chia-Jeng Chen. Meaningful trend in climate time series: A discussion based on linear and smoothing techniques for drought analysis in taiwan. Atmosphere, 13(3), 2022

    work page 2022

  24. [24]

    Unsupervised feature extraction by time- contrastive learning and nonlinear ica

    Aapo Hyvarinen and Hiroshi Morioka. Unsupervised feature extraction by time- contrastive learning and nonlinear ica. In Advances in Neural Information Processing Systems, pages 3765–3773, 2016

    work page 2016

  25. [25]

    Independent component analysis: algorithms and applications

    Aapo Hyv¨ arinen and Erkki Oja. Independent component analysis: algorithms and applications. Neural networks, 13(4-5):411–430, 2000

    work page 2000

  26. [26]

    Nonlinear ica using auxiliary variables and generalized contrastive learning

    Aapo Hyvarinen, Hiroaki Sasaki, and Richard Turner. Nonlinear ica using auxiliary variables and generalized contrastive learning. In The 22nd International Conference on Artificial Intelligence and Statistics , pages 859–868. PMLR, 2019

    work page 2019

  27. [27]

    Random survival forests

    Hemant Ishwaran, Udaya B Kogalur, Eugene H Blackstone, Michael S Lauer, et al. Random survival forests. The annals of applied statistics , 2(3):841–860, 2008

    work page 2008

  28. [28]

    A deep survival analysis method based on ranking

    Bingzhong Jing, Tao Zhang, Zixian Wang, Ying Jin, Kuiyuan Liu, Wenze Qiu, Liangru Ke, Ying Sun, Caisheng He, Dan Hou, et al. A deep survival analysis method based on ranking. Artificial intelligence in medicine , 98:1–9, 2019

    work page 2019

  29. [29]

    Deepsurv: personalized treatment recommender system using a cox proportional hazards deep neural network

    Jared L Katzman, Uri Shaham, Alexander Cloninger, Jonathan Bates, Tingting Jiang, and Yuval Kluger. Deepsurv: personalized treatment recommender system using a cox proportional hazards deep neural network. BMC medical research methodology , 18(1):24, 2018

    work page 2018

  30. [30]

    Maurice G. Kendall. Rank Correlation Methods. 4th edition edition, 1975. 11

    work page 1975

  31. [31]

    Adam: A Method for Stochastic Optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 , 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  32. [32]

    Analysis of soil moisture trends in europe using rank-based and empirical decomposition approaches

    Almendra-Mart´ ın Laura, Mart´ ınez-Fern´ andez Jos´ e, Piles Mar´ ıa, Gonz´ alez-Zamora ´Angel, Benito-Verdugo Pilar, and Gaona Jaime. Analysis of soil moisture trends in europe using rank-based and empirical decomposition approaches. Global and Plane- tary Change, 215:103868, 2022

    work page 2022

  33. [33]

    Contrastive representation learning: A framework and review

    Phuc H Le-Khac, Graham Healy, and Alan F Smeaton. Contrastive representation learning: A framework and review. IEEE Access, 2020

    work page 2020

  34. [34]

    Henry B. Mann. Nonparametric tests against trend. Econometrica, 13(3):245–259, 1945

    work page 1945

  35. [35]

    Survival analysis

    Rupert G Miller Jr. Survival analysis. John Wiley & Sons, 2011

    work page 2011

  36. [36]

    Techniques of trend analysis in degradation-based prognostics

    Seyed A Niknam, John Kobza, and J Wesley Hines. Techniques of trend analysis in degradation-based prognostics. The International Journal of Advanced Manufacturing Technology, 88(9-12):2429–2441, 2017

    work page 2017

  37. [37]

    Time series source sep- aration with slow flows

    Edouard Pineau, S´ ebastien Razakarivony, and Thomas Bonald. Time series source sep- aration with slow flows. In ICML Workshop on Invertible Neural Networks, Normalizing Flows, and Explicit Likelihood Models , 2020

    work page 2020

  38. [38]

    Unsupervised ageing detection of mechanical systems on a causality graph

    Edouard Pineau, S´ ebastien Razakarivony, and Thomas Bonald. Unsupervised ageing detection of mechanical systems on a causality graph. In ICMLA, 2020

    work page 2020

  39. [39]

    Robust contrastive learning and nonlinear ica in the presence of outliers

    Hiroaki Sasaki, Takashi Takenouchi, Ricardo Monti, and Aapo Hyvarinen. Robust contrastive learning and nonlinear ica in the presence of outliers. In Conference on Uncertainty in Artificial Intelligence , pages 659–668. PMLR, 2020

    work page 2020

  40. [40]

    Turbofan engine degradation simulation data set

    A Saxena and K Goebel. Turbofan engine degradation simulation data set. NASA Ames Prognostics Data Repository, 2008

    work page 2008

  41. [41]

    Hlynsson, and Laurenz Wiskott

    Merlin Sch¨ uler, Hlynur D. Hlynsson, and Laurenz Wiskott. Gradient-based training of slow feature analysis by differentiable approximate whitening. In Asian Conference on Machine Learning, pages 316–331. PMLR, 2019

    work page 2019

  42. [42]

    Randomized 2 x 2 trial evaluating hormonal treatment and the duration of chemotherapy in node-positive breast cancer patients

    M Schumacher, G Bastert, H Bojar, K Huebner, M Olschewski, W Sauerbrei, C Schmoor, C Beyerle, RL Neumann, and HF Rauschecker. Randomized 2 x 2 trial evaluating hormonal treatment and the duration of chemotherapy in node-positive breast cancer patients. german breast cancer study group. Journal of Clinical On- cology, 12(10):2086–2093, 1994

    work page 2086

  43. [43]

    On ranking in survival analysis: Bounds on the concordance index

    Harald Steck, Balaji Krishnapuram, Cary Dehing-Oberije, Philippe Lambin, and Vikas C Raykar. On ranking in survival analysis: Bounds on the concordance index. In Advances in neural information processing systems , pages 1209–1216, 2008

    work page 2008

  44. [44]

    Likelihood-free inference by ratio estimation

    Owen Thomas, Ritabrata Dutta, Jukka Corander, Samuel Kaski, and Michael U Gut- mann. Likelihood-free inference by ratio estimation. arXiv preprint arXiv:1611.10242, 2016

    work page arXiv 2016

  45. [45]

    Unsupervised learning of visual representations using videos

    Xiaolong Wang and Abhinav Gupta. Unsupervised learning of visual representations using videos. In Proceedings of the IEEE International Conference on Computer Vision, pages 2794–2802, 2015

    work page 2015

  46. [46]

    Vegetation dynamic trends and the main drivers detected using the en- semble empirical mode decomposition method in east africa

    Fangli Wei, Shuai Wang, Bojie Fu, Naiqing Pan, Xiaoming Feng, Wenwu Zhao, and Cong Wang. Vegetation dynamic trends and the main drivers detected using the en- semble empirical mode decomposition method in east africa. Land Degradation & Development, 29(8):2542–2553, 2018

    work page 2018

  47. [47]

    Slow feature analysis: Unsupervised learn- ing of invariances

    Laurenz Wiskott and Terrence J Sejnowski. Slow feature analysis: Unsupervised learn- ing of invariances. Neural computation, 14(4):715–770, 2002

    work page 2002

  48. [48]

    Learning patient- specific cancer survival distributions as a sequence of dependent regressors

    Chun-Nam Yu, Russell Greiner, Hsiu-Chin Lin, and Vickie Baracos. Learning patient- specific cancer survival distributions as a sequence of dependent regressors. In Advances in Neural Information Processing Systems , pages 1845–1853, 2011

    work page 2011

  49. [49]

    Serial-emd: Fast empirical mode decomposition method for multi- dimensional signals based on serialization

    Jin Zhang, Fan Feng, Pere Marti-Puig, Cesar F Caiafa, Zhe Sun, Feng Duan, and Jordi Sol´ e-Casals. Serial-emd: Fast empirical mode decomposition method for multi- dimensional signals based on serialization. Information Sciences, 581:215–232, 2021. 12

    work page 2021