Pith. sign in

REVIEW 3 major objections 5 minor 57 references

Neural Total Variation Distance Estimators for Changepoint Detection in News Data

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that a classifier trained to separate adjacent time windows of news articles yields an estimate of the total variation distance between their content distributions, with peaks marking 9/11, the 2016 US election, and the…

desk verdict Sound learning-by-confusion application, but the quantitative TV claim depends on an in-sample error estimate the paper never explicitly clears up. read the letter →

arxiv 2506.18764 v1 pith:JPFDVVCP submitted 2025-06-23 cs.LG cs.CLcs.CYcs.SI

Csaba Zsolnai , Niels Lörch , Julian Arnold This is my paper · ORCID
classification cs.LGcs.CLcs.CYcs.SI
keywords changepointdetectiontotalvariationdistancelearning-by-confusionnewsdiscourseneuralclassifierstransformerembeddingspublicshiftsmultitaskclassification
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to detect when public discourse changes by measuring, for each candidate date, how distinguishable the news articles before that date are from the articles after it. It claims that a classifier's error rate in separating the two windows estimates the total variation distance between the underlying content distributions, so large estimated distances mark changepoints. On a substantial real-world news corpus, the paper reports that this score peaks at events such as 9/11, the 2016 US election, and the Wuhan lockdown, and that it outperforms a topic-model baseline on the real data. The approach matters because it needs no hand-labeled changepoints, no topic taxonomy, and no domain-specific features.

What carries the argument

The load-bearing object is the learning-by-confusion score: for a candidate changepoint $t^*$, a binary classifier (here, multinomial logistic regression on transformer embeddings) is trained to distinguish articles in the window before $t^*$ from articles in the window after $t^*$. The classifier's estimated error rate $\hat{p}_{\mathrm{err}}(t^*)$ enters $\hat{D}_{\mathrm{TV}}(t^*) = 1 - 2\hat{p}_{\mathrm{err}}(t^*)$, which lower-bounds the true total variation distance between the two segment distributions. Training all candidate changepoints jointly through multi-task learning keeps the sweep computationally feasible, and the resulting score curve is scanned for significant maxima to identify changepoints.

What would settle it

Re-run the Guardian analysis with the same 80/20 date-based split, but compute the error rate only on the 20% held-out articles of each segment; if the AUC drops from 0.892 toward chance, or if the peaks for 9/11, the 2016 election, and the Wuhan lockdown disappear, the reported signal is a training-set artifact rather than a measure of content change.

Watch

Extended reading notes

Core claim

The paper's central claim is that the total variation distance between the content distributions of two consecutive time segments of news articles can be estimated from the error rate of a classifier trained to tell the segments apart. Using the identity $D_{\mathrm{TV}}[p,q] = 1 - 2 p_{\mathrm{err}}^{\mathrm{opt}}$, where $p_{\mathrm{err}}^{\mathrm{opt}}$ is the Bayes-optimal error in binary hypothesis testing, the authors argue that any classifier's estimated error rate $\hat{p}_{\mathrm{err}}$ gives a lower-bound estimate $\hat{D}_{\mathrm{TV}} = 1 - 2\hat{p}_{\mathrm{err}}$. On real news data, they report that this quantity's peaks align with major historical events and that the confusion-based method outperforms a latent-Dirichlet-allocation topic-model baseline at matching those events, with an AUC of 0.892 versus 0.839.

Load-bearing premise

The computation of the error rate in Eq. (3) treats the segment datasets as if the classifier's mistakes there are measured on data it never trained on, but the paper only describes an 80/20 split for early stopping and never states that the reported accuracy is out-of-sample; if the accuracy is in-sample, the estimated total variation distance is inflated by overfitting and the peaks may not reflect true content changes.

Editorial extensions

If this is right

  • On the real-world corpus of 197,170 news articles spanning 2000 to 2022, the confusion score produces peaks near 9/11, the 2016 US election, and the Wuhan lockdown, locating shifts the method was never told about.
  • The method's area under the curve on real news data is 0.892, beating the 0.839 of the topic-model baseline, meaning it matches a historical-event list more often at every tolerance window.
  • Because the score is a continuous quantity rather than a binary flag, peaks can be ranked by magnitude, which suggests a practical workflow: let the largest estimated distances flag which candidate dates deserve human inspection.
  • No labeled changepoints, topic lists, or hand-crafted features are required, so the same pipeline can be run on any date-stamped text corpus to generate a candidate timeline of public-discourse shifts.
  • The estimator's agnosticism to the article representation means swapping the embedding model changes the measured shifts without changing the algorithm, giving a direct route to adapting the method to other languages or text genres.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the method relies only on a classifier's error rate, the same multitask sweep could be turned into a streaming alarm: as each new day arrives, only a forward pass through the classifier is needed to update the score and flag a threshold crossing.
  • The identity between classification error and total variation distance suggests that replacing the binary head with a calibrated density-ratio head would let the same training machinery estimate other f-divergences, such as Kullback-Leibler or chi-squared, giving users a choice of which notion of change to prioritize.
  • The paper's benchmark results show LDA slightly ahead on the two synthetic tasks while the confusion method wins on real news; a natural follow-up is to synthesize datasets with shifts that are partly topical and partly tonal, testing where the crossover between the two methods lies.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper introduces a changepoint detection method for news text based on the learning-by-confusion scheme. For each candidate time t*, a classifier is trained to distinguish articles from the two adjacent time segments, and its error rate p_hat_err is converted into an estimated total variation distance via D_hat_TV = 1 - 2 p_hat_err. Changepoints are identified as peaks of this score. The method is evaluated on two synthetic/controlled benchmarks and on 27 yearly Guardian datasets, comparing against an LDA topic-distance baseline and a random baseline. The authors report that the confusion method outperforms LDA on the real Guardian data (AUC 0.892 vs 0.839) and that both methods detect events such as 9/11, the 2016 U.S. election, and the Wuhan lockdown.

Significance. The conceptual connection between classification error and total variation distance is classical and elegant, and adapting learning-by-confusion to high-dimensional text via transformer embeddings is a plausible and potentially useful approach. The paper ships a GitHub repository, and the benchmarks with induced changepoints provide a reproducible sanity check. If the quantitative TV estimate were honestly computed on held-out data, the method would offer an interpretable dissimilarity score for high-dimensional, noisy text where density estimation is infeasible. The real-data comparison with LDA is also of practical interest. However, the central quantitative claim currently rests on an error estimate whose evaluation protocol is not specified, which is a load-bearing gap.

major comments (3)
  1. [Sec. 2.2, Eq. (3); Sec. 2.3] The error rate in Eq. (3) is defined by averaging over the same segment datasets D_y(t*) that are used to minimize the loss in Eq. (2). The text nowhere states that p_hat_err is evaluated on held-out articles. Section 2.3 introduces an 80/20 train/validation split, but only for early stopping ('we stop the training if the validation loss stops improving'); validation loss is not the reported error. Consequently, D_hat_TV = 1 - 2 p_hat_err is not established as a lower-bound estimate of the true TV distance: in-sample error is biased downward by overfitting, so D_hat_TV can exceed D_TV. The risk is concrete because the classifier is logistic regression on 768-dimensional DistilRoBERTa embeddings and Benchmark 1 segments contain only 80 articles per side. Please specify explicitly whether the reported p_hat_err is computed on the validation split, and if not, recompute all reported TV scores and the resulting AUC/detection metrics using held-out predictions (e.g., validation-fold or cross-validated error). This is necessary to support the abstract's claim of a 'quantitative measure of change in content' and the discussion in Sec. 7.
  2. [Sec. 3.3 and App. B] Hyperparameter selection with Optuna is described as optimizing 'the objective of minimizing the ∆ metric,' but the text does not state whether this optimization is performed on a separate validation set or through nested cross-validation. If the hyperparameters for both the confusion method and the LDA baseline are selected using the same datasets on which the final ∆ and AUC are then reported, the reported performance metrics (including the headline Guardian AUC comparison 0.892 vs 0.839) are optimistically biased. Please clarify the selection protocol and, if needed, use a held-out protocol so that the empirical claims in Table 1 are not based on test-data fitting.
  3. [Sec. 2.2, asymptotic claim] The sentence 'In the infinite-data limit, the optimal classifier under the loss in Eq. (2) attains the Bayes optimal error rate' is not justified for the specific model class used, namely linear logistic regression on a fixed pretrained embedding. For the estimator to recover D_TV asymptotically, the model class must be flexible enough to represent the Bayes decision boundary. The paper should either qualify this statement to refer to a universally consistent classifier family or state the approximation incurred by the linear model on fixed embeddings.
minor comments (5)
  1. [Table 1] The table formatting appears corrupted: entries such as '80.50±0.08', '1801.1±0.2', and '18059±4' likely represent separate L and Δ columns (e.g., '8 | 0.50±0.08', '180 | 1.1±0.2', '180 | 59±4'). Please reformat the table so that the L column and the Δ column are clearly separated.
  2. [Eq. (1)] Equation (1) writes the total variation distance as a sum over x∈X, which is only well-defined for discrete X. Since the text also refers to probability distributions over articles, please clarify whether X is treated as discrete or replace the sum with the appropriate supremum/integral formulation.
  3. [Footnote 1] The claim that TV distance lower-bounds 'various other statistical distances' is imprecise; for many f-divergences the relationship involves a constant factor or is not a uniform lower bound. Please state the precise inequalities with references.
  4. [Sec. 2.3] There is a minor typo: 'Each training runs for at least 1000 epochs' should be 'Each training run lasts for at least 1000 epochs.'
  5. [Sec. 4] The text says the SE on the AUC metric is omitted because it is negligible, but the table header does not indicate what 'SE' refers to in the table cells. Please add a footnote defining the reported error bars consistently across tables and figures.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the TV-distance estimate rests on a classical Bayes-error identity and external ground truth; only minor author-overlap citations and an ambiguous evaluation split are noted.

full rationale

The derivation is self-contained rather than circular. D_TV is defined independently in Eq. (1); the identity D_TV = 1 - 2 p_opt_err cited to [28] is a classical, parameter-free fact that can be verified independently of the authors' prior work, so the author overlap in that citation is not load-bearing. The multi-task variant [27] is an efficiency device, not the source of the quantitative claim. Both benchmarks use externally specified ground truth (GPT-induced topic changes, category switches, and onthisday.com events), so no reported AUC or delta value is a fitted parameter renamed as a prediction. The one passage that warrants attention is Sec. 2.2: Eq. (3) computes p_hat_err over the same segment datasets D_y(t*) used in the loss Eq. (2), and Sec. 2.3 mentions the 80/20 validation split only for early stopping, never explicitly stating that p_hat_err is evaluated on held-out articles. If Eq. (3) were in-sample, D_hat_TV = 1 - 2 p_hat_err would be a monotone transform of training accuracy rather than an independent estimate of D_TV. However, this is an evaluation-validity ambiguity, not a demonstrated definitional reduction: the text does not say the reported error is the training error, and the identity itself is not built from the fitted values. The self-citations are present but not load-bearing; no step of the derivation is equivalent to its inputs by construction.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on no new physical or mathematical entities. What it does consume: one time-scale parameter L chosen per dataset after seeing performance, a set of training hyperparameters and embedding model selected by Optuna on the benchmark test sets, baseline LDA hyperparameters tuned the same way, the standard Bayes-error identity, an unstated reliance on embedding sufficiency, the i.i.d. assumption, and an event-list proxy for ground truth. The Optuna selections make the reported benchmark numbers optimistic; the embedding sufficiency and i.i.d. assumptions are the substantive domain premises.

free parameters (3)
  • L (segment half-width in days) = 8 (30-day benchmark), 180 (yearly sets)
    Sets the time scale of comparison; chosen by 'empirical observation' and tuned via Optuna to minimize the deviation metric on the target datasets (Sec. 4, App. B), so reported scores are on the tuning data.
  • Embedding model, learning rate, batch size (confusion method) = all-distilroberta-v1, 8e-5, 64
    Selected by Optuna with the objective of minimizing the deviation metric on the benchmark datasets (Table B1), which biases the benchmark comparison.
  • LDA baseline hyperparameters (topics, iterations, passes, chunk size) = 20, 83, 22, 319
    Tuned by Optuna on the same benchmarks (Table B2); relevant because the headline claim is beating the LDA baseline.
assumptions (4)
  • standard math Bayes-error identity: D_TV[p,q] = 1 - 2 p_opt_err, and the cross-entropy-trained classifier approaches the Bayes error in the infinite-data limit.
    Invoked in Sec. 2.2 to convert classification accuracy into a TV distance estimate; the identity is classical, though the paper cites its own ref. [28].
  • domain assumption The Distilroberta embedding preserves enough information that the TV distance between embedded distributions tracks the TV distance between article content distributions.
    Introduced in Sec. 2.3; a semantic shift invisible to the sentence embedding is invisible to the method, so the 'quantitative measure of change in content' is only as faithful as the embedding.
  • domain assumption Articles within a time segment are independent and identically distributed.
    Stated in Sec. 2.1 and acknowledged in Sec. 6 as an explicit limitation; serial correlation in news coverage can produce false detection signals.
  • domain assumption onthisday.com historical events are a usable proxy for ground-truth changepoints in The Guardian coverage.
    Used in Sec. 3.3 for the real-data evaluation; the authors note coverage mismatch and temporal misalignment in Sec. 6.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Neural Total Variation Distance Estimators for Changepoint Detection in News Data." pith.science (2026). https://pith.science/paper/JPFDVVCP

@misc{pith2026250618764,
  author       = {Pith},
  title        = {Pith review of: Neural Total Variation Distance Estimators for Changepoint Detection in News Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JPFDVVCP}},
  note         = {Machine review of arXiv:2506.18764}
}
read the original abstract

Detecting when public discourse shifts in response to major events is crucial for understanding societal dynamics. Real-world data is high-dimensional, sparse, and noisy, making changepoint detection in this domain a challenging endeavor. In this paper, we leverage neural networks for changepoint detection in news data, introducing a method based on the so-called learning-by-confusion scheme, which was originally developed for detecting phase transitions in physical systems. We train classifiers to distinguish between articles from different time periods. The resulting classification accuracy is used to estimate the total variation distance between underlying content distributions, where significant distances highlight changepoints. We demonstrate the effectiveness of this method on both synthetic datasets and real-world data from The Guardian newspaper, successfully identifying major historical events including 9/11, the COVID-19 pandemic, and presidential elections. Our approach requires minimal domain knowledge, can autonomously discover significant shifts in public discourse, and yields a quantitative measure of change in content, making it valuable for journalism, policy analysis, and crisis monitoring.

Figures

Figures reproduced from arXiv: 2506.18764 by the authors.

Figure 1
Figure 1. Illustration of the problem setup for detecting changepoints in news articles. Here, we [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Dissimilarity scores for an example from the benchmark 1 dataset (30-day interval) [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Success rate as a function of threshold n for the Guardian dataset (L = 180). Here, the SE is negligible (< 0.01 on average) and hence we omit it. While the confusion-based method did not provide any advantage over the LDA-based baseline method in the case of the well-structured benchmark datasets, it excels at capturing the changepoints in real-world news data. This suggests that the added complexity of our approac… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Dissimilarity score curves for different examples from the Guardian dataset computed via [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

57 extracted references · 49 canonical work pages

  1. [1]

    Real-time financial surveillance via quickest change-point detection methods.arXiv:1509.01570, 2015

    Andrey Pepelyshev and Aleksey S Polunchenko. Real-time financial surveillance via quickest change-point detection methods.arXiv:1509.01570, 2015

    arXiv 2015

  2. [2]

    Detection of uterine MMG contractions using a multiple change point estimator and the K-means cluster algorithm.IEEE Trans

    Patricio S La Rosa, Arye Nehorai, Hari Eswaran, Curtis L Lowery, and Hubert Preissl. Detection of uterine MMG contractions using a multiple change point estimator and the K-means cluster algorithm.IEEE Trans. Biomed. Eng., 55(2):453–467, 2008

    work page 2008

  3. [3]

    Online Bayesian change point detection algorithms for segmentation of epileptic activity

    Rakesh Malladi, Giridhar P Kalamangalam, and Behnaam Aazhang. Online Bayesian change point detection algorithms for segmentation of epileptic activity. In2013 Asilomar Conference on Signals, Systems and Computers, pages 1833–1837. IEEE, 2013

    work page 2013

  4. [4]

    A review and comparison of changepoint detection techniques for climate data.J

    Jaxk Reeves, Jien Chen, Xiaolan L Wang, Robert Lund, and Qi Qi Lu. A review and comparison of changepoint detection techniques for climate data.J. Appl. Meteorol. Climatol., 46(6): 900–915, 2007

    work page 2007

  5. [5]

    Change-point detection of climate time series by nonparametric method

    Naoki Itoh and Jürgen Kurths. Change-point detection of climate time series by nonparametric method. InProceedings of the World Congress on Engineering and Computer Science, volume 1, pages 445–448, 2010

    work page 2010

  6. [6]

    A changepoint approach to modelling non-stationary soil moisture dynamics

    Mengyi Gong, Rebecca Killick, Christopher Nemeth, and John Quinton. A changepoint approach to modelling non-stationary soil moisture dynamics.arXiv:2310.17546, 2023

    work page Pith review arXiv 2023

  7. [7]

    Is a Recent Surge in Global Warming Detectable?

    Claudie Beaulieu, Colin Gallagher, Rebecca Killick, Robert Lund, and Xueheng Shi. Is a recent surge in global warming detectable?arXiv:2403.03388, 2024

    work page Pith review arXiv 2024

  8. [8]

    Audio segmentation for speech recognition using segment features

    David Rybach, Christian Gollan, Ralf Schluter, and Hermann Ney. Audio segmentation for speech recognition using segment features. In2009 IEEE International Conference on Acoustics, Speech and Signal Processing, pages 4197–4200. IEEE, 2009

    work page 2009

Show all 57 references
  1. [9]

    Speaker change point detection using deep neural nets

    Vishwa Gupta. Speaker change point detection using deep neural nets. In2015 IEEE Interna- tional Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 4420–4424. IEEE, 2015

    2015

  2. [10]

    Statistically significant detection of linguistic change

    Vivek Kulkarni, Rami Al-Rfou, Bryan Perozzi, and Steven Skiena. Statistically significant detection of linguistic change. InProceedings of the 24th International Conference on World Wide Web, pages 625–635, 2015

    2015

  3. [11]

    A survey of methods for time series change point detection.Knowl

    Samaneh Aminikhanghahi and Diane J Cook. A survey of methods for time series change point detection.Knowl. Inf. Syst., 51(2):339–367, 2017

    2017

  4. [12]

    An evaluation of change point detection algorithms.arXiv:2003.06222, 2020

    Gerrit JJ Van den Burg and Christopher KI Williams. An evaluation of change point detection algorithms.arXiv:2003.06222, 2020

    2003 arXiv

  5. [13]

    Selective review of offline change point detection methods.Signal Process., 167:107299, 2020

    Charles Truong, Laurent Oudre, and Nicolas Vayatis. Selective review of offline change point detection methods.Signal Process., 167:107299, 2020

    2020

  6. [14]

    High-dimensional changepoint detection via a geometrically inspired mapping.Stat

    Thomas Grundy, Rebecca Killick, and Gueorgui Mihaylov. High-dimensional changepoint detection via a geometrically inspired mapping.Stat. Comput., 30(4):1155–1166, 2020

    2020

  7. [15]

    High-dimensional change-point detection using generalized homogeneity metrics.arXiv:2105.08976, 2021

    Shubhadeep Chakraborty and Xianyang Zhang. High-dimensional change-point detection using generalized homogeneity metrics.arXiv:2105.08976, 2021

    2021 arXiv

  8. [16]

    Anomaly detection using the Kullback- Leibler divergence metric

    Mostafa Afgani, Sinan Sinanovic, and Harald Haas. Anomaly detection using the Kullback- Leibler divergence metric. In2008 First International Symposium on Applied Sciences on Biomedical and Communication Technologies, pages 1–5. IEEE, 2008

    2008

  9. [17]

    Tracking changes using Kullback-Leibler divergence for the continual learning

    Sebastián Basterrech and Michał Wo´ zniak. Tracking changes using Kullback-Leibler divergence for the continual learning. In2022 IEEE International Conference on Systems, Man, and Cybernetics (SMC), pages 3279–3285. IEEE, 2022

    2022

  10. [18]

    Automatic segmentation, classification and clustering of broadcast news audio

    Matthew A Siegler, Uday Jain, Bhiksha Raj, and Richard M Stern. Automatic segmentation, classification and clustering of broadcast news audio. InProc. DARPA speech recognition workshop, volume 1997, 1997. 10

    1997

  11. [19]

    Multispectral change detection using multivariate Kullback-Leibler distance.ISPRS J

    Shabnam Jabari, Mohammad Rezaee, Fatemeh Fathollahi, and Yun Zhang. Multispectral change detection using multivariate Kullback-Leibler distance.ISPRS J. Photogramm. Remote Sens., 147:163–177, 2019

    2019

  12. [20]

    Direct importance estimation for covariate shift adaptation.Ann

    Masashi Sugiyama, Taiji Suzuki, Shinichi Nakajima, Hisashi Kashima, Paul V on Bünau, and Motoaki Kawanabe. Direct importance estimation for covariate shift adaptation.Ann. Inst. Stat. Math., 60:699–746, 2008

    2008

  13. [21]

    Change-point detection in time-series data by direct density-ratio estimation

    Yoshinobu Kawahara and Masashi Sugiyama. Change-point detection in time-series data by direct density-ratio estimation. InProceedings of the 2009 SIAM International Conference on Data Mining (SDM), pages 389–400. SIAM, 2009

    2009

  14. [22]

    Change-point detection in time-series data by relative density-ratio estimation.Neural Netw., 43:72–83, 2013

    Song Liu, Makoto Yamada, Nigel Collier, and Masashi Sugiyama. Change-point detection in time-series data by relative density-ratio estimation.Neural Netw., 43:72–83, 2013

    2013

  15. [23]

    Efficient multistream classification using direct density ratio estimation

    Ahsanul Haque, Swarup Chandra, Latifur Khan, Kevin Hamlen, and Charu Aggarwal. Efficient multistream classification using direct density ratio estimation. In2017 IEEE 33rd International Conference on Data Engineering (ICDE), pages 155–158, 2017

    2017

  16. [24]

    Learning phase transitions by confusion.Nature Phys., 13(5):435–439, 2017

    Evert PL Van Nieuwenburg, Ye-Hua Liu, and Sebastian D Huber. Learning phase transitions by confusion.Nature Phys., 13(5):435–439, 2017

    2017

  17. [25]

    Replacing neural networks by optimal analytical predictors for the detection of phase transitions.Phys

    Julian Arnold and Frank Schäfer. Replacing neural networks by optimal analytical predictors for the detection of phase transitions.Phys. Rev. X, 12:031044, Sep 2022

    2022

  18. [26]

    Mapping out phase diagrams with generative classifiers.Phys

    Julian Arnold, Frank Schäfer, Alan Edelman, and Christoph Bruder. Mapping out phase diagrams with generative classifiers.Phys. Rev. Lett., 132:207301, May 2024

    2024

  19. [27]

    Fast detection of phase transitions with multi-task learning-by-confusion.arXiv:2311.09128, 2023

    Julian Arnold, Frank Schäfer, and Niels Lörch. Fast detection of phase transitions with multi-task learning-by-confusion.arXiv:2311.09128, 2023

    2023 arXiv

  20. [28]

    Machine learning phase transitions: Connections to the Fisher information.arXiv:2311.10710, 2023

    Julian Arnold, Niels Lörch, Flemming Holtorf, and Frank Schäfer. Machine learning phase transitions: Connections to the Fisher information.arXiv:2311.10710, 2023

    2023 arXiv

  21. [29]

    Quantum chi-squared tomography and mutual infor- mation testing.Quantum, 8:1381, 2024

    Steven T Flammia and Ryan O’Donnell. Quantum chi-squared tomography and mutual infor- mation testing.Quantum, 8:1381, 2024

    2024

  22. [30]

    Neural changepoint detection in news data: A learning-by-confusion approach for identifying shifts in public dis- course

    Csaba Zsolnai, Niels Lörch, and Julian Arnold. GitHub repository for “Neural changepoint detection in news data: A learning-by-confusion approach for identifying shifts in public dis- course”.https://github.com/csabazsolnai/news_changepoint_detection, 2025

    2025

  23. [31]

    DistilRoBERTa Base Sentence Transformer, 2022

    Hugging Face Model Hub. DistilRoBERTa Base Sentence Transformer, 2022. Accessed on 23 May 2025

    2022

  24. [32]

    Term-weighting approaches in automatic text retrieval

    Gerard Salton and Christopher Buckley. Term-weighting approaches in automatic text retrieval. Inf. Process. Manag., 24(5):513–523, 1988

    1988

  25. [33]

    Adam: A method for stochastic optimization

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

    2014 arXiv

  26. [34]

    Optuna: A next-generation hyperparameter optimization framework

    Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. Optuna: A next-generation hyperparameter optimization framework. InProceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2019

    2019

  27. [35]

    Latent Dirichlet allocation.J

    David M Blei, Andrew Y Ng, and Michael I Jordan. Latent Dirichlet allocation.J. Mach. Learn. Res., 3:993–1022, 2003

    2003

  28. [36]

    Change point detection in time series data using autoencoders with a time-invariant representation.IEEE Trans

    Tim De Ryck, Maarten De V os, and Alexander Bertrand. Change point detection in time series data using autoencoders with a time-invariant representation.IEEE Trans. Signal Process., 69: 3513–3524, 2021

    2021

  29. [37]

    Time series change point detec- tion with self-supervised contrastive predictive coding

    Shohreh Deldari, Daniel V Smith, Hao Xue, and Flora D Salim. Time series change point detec- tion with self-supervised contrastive predictive coding. InProceedings of the Web Conference 2021, pages 3124–3135, 2021. 11

    2021

  30. [38]

    Deep learning for multi-scale changepoint detection in multivariate time series.arXiv:1905.06913, 2019

    Zahra Ebrahimzadeh, Min Zheng, Selcuk Karakas, and Samantha Kleinberg. Deep learning for multi-scale changepoint detection in multivariate time series.arXiv:1905.06913, 2019

    1905 arXiv

  31. [39]

    Zero-shot speaker change point detection using large language models

    Yannis Tevissen, Jérôme Boudy, Gérard Chollet, and Frédéric Petitpont. Zero-shot speaker change point detection using large language models. InJournée des doctorants Paris Saclay, 2023

    2023

  32. [40]

    Deep density ratio estimation for change point detection.arXiv:1905.09876, 2019

    Haidar Khan, Lara Marcuse, and Bülent Yener. Deep density ratio estimation for change point detection.arXiv:1905.09876, 2019

    1905 arXiv

  33. [41]

    Training neural networks for likelihood/density ratio estimation.arXiv:1911.00405, 2019

    George V Moustakides and Kalliopi Basioti. Training neural networks for likelihood/density ratio estimation.arXiv:1911.00405, 2019

    1911 arXiv

  34. [42]

    Continual density ratio estimation in an online setting.arXiv:2103.05276, 2021

    Yu Chen, Song Liu, Tom Diethe, and Peter Flach. Continual density ratio estimation in an online setting.arXiv:2103.05276, 2021

    2021 arXiv

  35. [43]

    Inferences for case-control and semiparametric two-sample density ratio models

    Jing Qin. Inferences for case-control and semiparametric two-sample density ratio models. Biometrika, 85(3):619–630, 1998

    1998

  36. [44]

    Semiparametric density estimation under a two-sample density ratio model.Bernoulli, 10(4):583–604, 2004

    Kuang Fu Cheng and Chih-Kang Chu. Semiparametric density estimation under a two-sample density ratio model.Bernoulli, 10(4):583–604, 2004

    2004

  37. [45]

    Discriminative learning under covariate shift.J

    Steffen Bickel, Michael Brückner, and Tobias Scheffer. Discriminative learning under covariate shift.J. Mach. Learn. Res., 10(9), 2009

    2009

  38. [46]

    Cambridge University Press, 2012

    Masashi Sugiyama, Taiji Suzuki, and Takafumi Kanamori.Density Ratio Estimation in Machine Learning. Cambridge University Press, 2012

    2012

  39. [47]

    Linking losses for density ratio and class-probability estimation

    Aditya Menon and Cheng Soon Ong. Linking losses for density ratio and class-probability estimation. In Maria Florina Balcan and Kilian Q. Weinberger, editors,Proceedings of The 33rd International Conference on Machine Learning, volume 48, pages 304–313, New York, New York, USA...

    2016

  40. [48]

    Breaking the curse of horizon: Infinite-horizon off-policy estimation.Adv

    Qiang Liu, Lihong Li, Ziyang Tang, and Dengyong Zhou. Breaking the curse of horizon: Infinite-horizon off-policy estimation.Adv. Neural Inf. Process. Syst. – NeurIPS, 31, 2018. URL https://proceedings.neurips.cc/paper/2018/hash/ dda04f9d634145a9c68d5dfe53b21272-Abstract.html

    2018

  41. [49]

    Gutmann and Aapo Hyvärinen

    Michael U. Gutmann and Aapo Hyvärinen. Noise-contrastive estimation of unnormalized statistical models, with applications to natural image statistics.Journal of Machine Learning Re- search, 13(11):307–361, 2012. URLhttp://jmlr.org/papers/v13/gutmann12a.html

    2012

  42. [50]

    f-GAN: Training Generative Neural Samplers using Variational Divergence Minimization

    Sebastian Nowozin, Botond Cseke, and Ryota Tomioka. f-GAN: Training Generative Neural Samplers using Variational Divergence Minimization. In D. Lee, M. Sugiyama, U. Luxburg, I. Guyon, and R. Garnett, editors,Advances in Neural Information Processing Systems, vol- ume 29. Curra...

    2016

  43. [51]

    Unsupervised change analysis using supervised learning

    Shohei Hido, Tsuyoshi Idé, Hisashi Kashima, Harunobu Kubo, and Hirofumi Matsuzawa. Unsupervised change analysis using supervised learning. InAdvances in Knowledge Discovery and Data Mining: 12th Pacific-Asia Conference, PAKDD 2008 Osaka, Japan, May 20-23, 2008 Proceedings 12, ...

    2008

  44. [52]

    Change point detection with neural online density-ratio estimator

    Xiuheng Wang, Ricardo Augusto Borsoi, Cédric Richard, and Jie Chen. Change point detection with neural online density-ratio estimator. InICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–5. IEEE, 2023

    2023

  45. [53]

    Leveraging change point detection to discover natural experiments in data.EPJ Data Sci., 11(1):49, 2022

    Yuzi He, Keith A Burghardt, and Kristina Lerman. Leveraging change point detection to discover natural experiments in data.EPJ Data Sci., 11(1):49, 2022

    2022

  46. [54]

    Discovering collective narratives shifts in online discussions

    Wanying Zhao, Siyi Guo, Kristina Lerman, and Yong-Yeol Ahn. Discovering collective narratives shifts in online discussions. InProceedings of the International AAAI Conference on Web and Social Media, volume 18, pages 1804–1817, 2024. 12

    2024

  47. [55]

    Benjamin Rhodes, Kai Xu, and Michael U. Gutmann. Telescoping Density-Ratio Estimation. In H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin, editors,Advances in Neural Information Processing Systems, volume 33, pages 4905–4916. Curran Associates, Inc., 2020

    2020

  48. [56]

    Featurized density ratio estimation

    Kristy Choi, Madeline Liao, and Stefano Ermon. Featurized density ratio estimation. In Cassio de Campos and Marloes H. Maathuis, editors,Proceedings of the Thirty-Seventh Conference on Uncertainty in Artificial Intelligence, volume 161 ofProceedings of Machine Learning Researc...

    2021

  49. [57]

    A comparison of single and multiple changepoint techniques for time series data.Comput

    Xuesheng Shi, Colin Gallagher, Robert Lund, and Rebecca Killick. A comparison of single and multiple changepoint techniques for time series data.Comput. Stat. Data Anal., 170:107433, 2022. 13 A Details on datasets Table A1 contains a few exemplary news article from The Guardia...

    2022

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.