Pith. sign in

REVIEW 4 major objections 6 minor 67 references

CORAL: Concept Drift Representation Learning for Co-evolving Time-series

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A kernel-induced self-representation matrix whose blocks track latent concepts can identify, track, and forecast concept drift in co-evolving time series.

desk verdict CORAL's cross-window transition model is a modest novelty, but its core derivation and Theorem 4.1 don't survive contact with the paper's own formulas. read the letter →

arxiv 2501.01480 v3 pith:BWUMGV25 submitted 2025-01-02 cs.LG cs.AI

Kunpeng Xu , Lifei Chen , Shengrui Wang This is my paper · ORCID
classification cs.LGcs.AI
keywords conceptdriftco-evolvingtimeseriesself-representationlearningkernelmethodsblock-diagonalrepresentationtemporaltrackingforecasting
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

CORAL claims that concept drift in co-evolving time series becomes visible as a changing block-diagonal structure in a learned representation matrix. For each sliding window, the paper solves a kernel self-representation problem whose regularizer drives the coefficient matrix Z toward k blocks, one per latent concept; tracking how these blocks appear, move, and disappear over windows both identifies drift and supports forecasting of future concepts and series values. The authors argue this makes drift detection adaptive and interpretable, requiring no prior knowledge of concepts, and demonstrate that the representation can be plugged into deep learning backbones. They also state the method's main limitation: with only a few series, the block regularizer loses effectiveness, so the approach targets wide co-evolving datasets.

What carries the argument

The central object is the kernel-induced self-representation matrix Z, together with the block-diagonal regularizer built from its Laplacian. Z is an n×n nonnegative symmetric matrix that reconstructs each series from the others after a kernel mapping into a reproducing kernel Hilbert space. The regularizer ||Z||_k sums the k smallest eigenvalues of the Laplacian L_Z; Theorem 4.1 ties this quantity to exact k-block diagonality. Around this object, CORAL builds a complete pipeline: an eigengap threshold on the Laplacian spectrum estimates the number of concepts, a spectral clustering step assigns series to blocks, a transition-probability rule blends each series' own concept history with the ecosystem's aggregate transition frequencies, and a recency-weighted forecast produces future values from the series' past under the predicted concept.

What would settle it

Run the optimization in Eq. 3 on a synthetic dataset with a known number of concepts from many initializations; if the resulting matrices retain sizable entries outside the diagonal blocks, or the count of near-zero Laplacian eigenvalues departs from the true concept count, the claimed block-diagonal equivalence does not hold.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the right representation turns concept drift into a visible, countable object. For each sliding window over an n-variable time series, CORAL learns an n×n matrix Z with nonnegative symmetric entries, expressing every series as a combination of the others in a kernel feature space; the ideal Z is block diagonal, with each diagonal block containing exactly the series that share a latent concept. Theorem 4.1 asserts that minimizing the sum of the k smallest eigenvalues of the Laplacian of Z is equivalent to Z being k-block diagonal, so the number of concepts can be read off from the zero eigenvalues. By re-solving the problem in consecutive windows, CORAL traces each series as it moves between blocks—this trajectory is the drift—and uses per-series transition counts combined with ecosystem-wide transition frequencies to forecast which concept a series will enter next and what its future values will be. The empirical sections show the block structure separating recognizable regimes (for example, calm versus volatile financial periods) and report forecasting errors at or near the best among the compared models on synthetic and real datasets.

Load-bearing premise

The load-bearing premise is that the mathematical penalty actually pushes the learned representation matrix toward clean, separated diagonal blocks; the paper proves that blocks produce zero penalty but leaves unproven that minimizing the penalty forces the optimization to produce blocks.

Editorial extensions

If this is right

  • The number of latent concepts can be estimated automatically from the eigengap of the Laplacian of Z, so CORAL requires no prior knowledge of how many concepts exist.
  • Drift becomes observable per series: by reading which block each series belongs to in each window, CORAL shows an individual series moving from one concept to another over time.
  • Forecasting uses both the series' own past concept transitions and transition frequencies across the whole ecosystem, letting it predict a concept that has never appeared in that series before.
  • Because the kernel representation layer can be inserted into autoencoder-style networks, deep learning backbones can gain drift-awareness without changing their training data.

Reading between the lines

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

  • A natural extension the paper does not develop is using the rate of block rearrangement as a drift-speed estimator: slow block shifts would indicate gradual drift and sudden reorderings abrupt regime breaks.
  • For datasets with few series, where the authors say the block regularizer weakens, a hierarchical prior or pairwise coupling across windows could restore structure; this is an inference, not a claim in the paper.
  • The transition-probability forecast could be run online, updating the ecosystem-level frequencies as each new window arrives, giving an anytime forecasting rule for streaming co-evolving series.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes CORAL, a method for concept drift identification and forecasting in co-evolving time series. For each sliding window, CORAL learns a nonnegative symmetric self-representation matrix Z of the series in a kernel-induced feature space, regularized by the sum of the k smallest eigenvalues of the graph Laplacian of Z. The authors claim that this regularizer enforces a k-block diagonal structure, which enables automatic discovery of the number k of concepts, tracking of concept drift through transition probabilities between blocks across windows, and forecasting of future concept values via a weighted average of past subseries. The paper evaluates the method on one synthetic dataset and several real datasets, comparing against both forecasting and concept-drift baselines.

Significance. If the central claims were sound, CORAL would provide an interpretable, unsupervised approach to regime discovery in high-dimensional co-evolving time series, with a matrix-optimization formulation and an explicit mathematical guarantee for concept-count recovery. The paper also demonstrates an extension to deep learning backbones and reports competitive or superior RMSE on several benchmarks. However, the load-bearing theoretical result, Theorem 4.1, is incorrect as stated; the algebraic expansion of the objective in Eq. (2) is inconsistent with the chosen parameter values; and the evaluation of concept detection on real financial data is not grounded in any defined ground truth. As a result, the validity of the paper's main claims is not established, and the empirical results cannot be taken as supporting the proposed mechanism.

major comments (4)
  1. [Sec. 4.1, Eq. (2)] The claimed expansion is algebraically incorrect. Expanding 1/2 ||Φ(S) − (α/2)Φ(S)Z||^2 gives 1/2 Tr(K) − (α/2)Tr(KZ) + (α^2/8)Tr(Z^T K Z). The printed right-hand side, 1/2 Tr(K − αKZ + Z^T K Z), has quadratic coefficient 1/2, which matches α^2/8 only for α=2. The paper later fixes α=4 (Appendix F), for which the printed quadratic coefficient is off by a factor of 4. The same inconsistency appears in Eq. (10), and the V-update in Eq. (15) uses αK rather than (α/2)K as required by Eq. (13). The exact objective being minimized is therefore ambiguous, and the reported regularization may not correspond to the stated loss.
  2. [Sec. 4.1, Theorem 4.1] The theorem claims that minimizing ||Z||_k = sum_{i=N-k+1}^N λ_i(L_Z) is equivalent to Z being k-block diagonal. The proof shows only that a block-diagonal Z with k blocks yields k zero eigenvalues. Since the Laplacian of any symmetric nonnegative Z with zero diagonal is positive semidefinite and its zero-eigenvalue multiplicity equals the number of connected components, the regularizer is zero for every Z with at least k connected components. The optimization in Eq. (3) is therefore not constrained to produce exactly k blocks, and the eigengap estimator of Eqs. (8)-(9) has no support: additional zero eigenvalues remove the gap used to determine k. This is the central guarantee for concept identification, so the main theoretical claim of the paper is not established.
  3. [Sec. G.3, Tables 5-6] The concept-detection evaluation reports TP/FP/TN/FN and Type I/II errors for Stock1 and Stock2, with a column labeled 'Ground Truth Concepts' equal to 5. However, Section 6.3 states that for real datasets 'we lack the ground truth for validating the obtained concepts.' No method is given for obtaining the financial ground-truth concepts. Without such a definition, the detection-accuracy numbers for these two datasets are uninterpretable.
  4. [Sec. 6.1, Table 1] The forecasting evaluation does not specify a clean train/validation/test split. The text says that a fixed window slides over all the series and that a kernel representation is learned for the subseries in each window; this suggests representations are learned on windows that are later used for evaluation. Eq. (6) forecasts a test window as a weighted average of observed subseries from past windows assigned to the predicted concept, so the extent to which the comparison reflects out-of-sample generalization is unclear. The paper should state exactly which windows are used for representation learning, for estimating transition probabilities, and for computing the reported RMSE.
minor comments (6)
  1. [Fig. 1 caption and Sec. 1] The caption and the accompanying text contain the stray string 'g♀be', which appears to be a rendering artifact; it should read 'ecosystem' or be removed.
  2. [Sec. 4.3, loss function] The displayed loss L(Θ) contains only λ1 and λ2, but the text says λ1, λ2, and λ3 balance the loss components. If λ3 was intended for a temporal smoothness term, that term is missing from the displayed equation.
  3. [Sec. 4.2, Eq. (6)] The condition sum_{l=1}^p τ^{p-l+1} = 1 is stated, but no specific form of τ is given; without specifying τ, the forecasting procedure is not fully reproducible.
  4. [Appendix B, Eq. (8)] The set-builder notation {i | g(σ_i) ≤ τ}_{i=1}^{N-1} is unusual and should be rewritten as a standard argmin over i ∈ {1,...,N-1} with the constraint g(σ_i) ≤ τ.
  5. [Supplementary title] The supplementary document title reads 'CORLA', which appears to be a typo for 'CORAL'.
  6. [Sec. 6.2] The claim that the exclusive appearance of C2 and C5 in the sixth window of Stock1 may be 'possibly induced by the COVID-19 pandemic' is presented without supporting evidence or a specific date; as a claim about financial regimes it should be stated more cautiously.

Circularity Check

2 steps flagged · score 6.0 of 10

The concept count is both an input to Eq. 3's k-block regularizer and the quantity "estimated" from the eigengaps of the resulting Z, and the Eq. 6 value forecast is a convex combination of the series' own past values selected by CORAL's own cluster labels, so the central discoveries are partially fixed by construction.

  1. self definitional [Sec. 4.1 (Eq. 3 and Theorem 4.1); Appendix B (Eqs. 8-9 and footnote 4)]
    "where γ>0 balances the loss function with regularization term, ||Z||_k = Σ_{i=N−k+1}^N λ_i(L_Z) ... Here, the regularization term is equal to 0 if and only if Z is k-block diagonal (see Theorem 4.1 for details). ... In this paper, we simply initialize the number of concepts for each window as 3 to obtain the initial representation Z, and then estimate k."

    The regularizer in Eq. 3 takes k as a free parameter: ||Z||_k is the sum of the k smallest Laplacian eigenvalues, and the paper states it is zero 'if and only if Z is k-block diagonal.' Minimizing it therefore manufactures (up to) k zero eigenvalues in L_Z. Appendix B then 'estimates' k by counting zero eigenvalues/eigengaps of L_Z of this very solution. The discovered concept count is the k fed into the objective, so the claimed autonomous identification of k is self-referential. The theorem's proof covers only the direction from k-block structure to k zero eigenvalues, never that the Eq. 3 minimizer has exactly k blocks, so the eigengap readout is not an independent data-driven identification.

  2. fitted input called prediction [Sec. 4.2 (Eq. 6); Sec. 6.3 validation protocol]
    "The predicted values of Si under the window Wp+1 can be calculated as: PreSi = Σ_{l=1}^p Δ(Cm|Si, Wl) · τ^{p−l+1} · {Si|Wl} ... In this context, we simply require Σ_{l=1}^p τ^{p−l+1} = 1, which implies that the subseries closer to the predicted window is deemed more significant. ... For real datasets, we lack the ground truth for validating the obtained concepts. Instead, we validate the value and gain of the discovered concepts for time series forecasting as they are employed in Eq. 6."

    By construction PreSi is a convex combination of the same series' observed subseries {Si|Wl}; the only fitted inputs are the concept labels Δ(Cm|Si,Wl) produced by CORAL's own clustering, and τ is a hand-set decay. Thus the 'forecast' is a label-conditioned moving average of the input windows, not a prediction generated from the learned kernel representation or from inter-series interactions. The paper then uses this forecast accuracy as the evidence for the discovered concepts, so the validation signal reduces to the model's own labels smoothing the data it was given.

full rationale

The only load-bearing circularity I can exhibit by the paper's own equations is the k-in/k-out loop. Eq. 3 asks for ||Z||_k, the sum of the k smallest Laplacian eigenvalues of Z, and the paper states this term is zero iff Z is k-block diagonal; hence solving Eq. 3 with an initialized k (Appendix B: 'we simply initialize the number of concepts for each window as 3') builds (at least) k zero eigenvalues into L_Z. Appendix B then counts zero eigenvalues/eigengaps of L_Z and presents that count as the discovered number of concepts. The discovered k is therefore the supplied k, not an independent quantity. The proof of Theorem 4.1 establishes only the direction from k-block structure to zero eigenvalues, never that the Eq. 3 minimizer has exactly k blocks, so the circular readout is also unsupported. Second, Eq. 6 defines the forecast for W_{p+1} as a weighted sum of {Si|Wl} over past windows, requiring the weights to sum to 1. This is, by construction, a convex combination of the same series' observed values, with the only model-dependent input being the cluster membership Δ(Cm|Si,Wl) produced by CORAL's own clustering. Using the accuracy of this forecast to validate the discovered concepts (Sec. 6.3) therefore validates the concepts through a smoothing of the data using those same concepts. I do not treat the self-citations as circular: the eigenvalue-reformulation property is also attributed to Dattorro (2010), an external source, and the earlier papers by the same group are used mainly for comparisons and motivation, not as the load-bearing justification. The algebraic inconsistency between Eq. 2's two displayed forms at α=4, and the similar issue in Eq. 10, are correctness defects that make the exact objective ambiguous; they are not themselves a reduction to inputs and do not affect the circularity score. External benchmarks (Table 1, Stock2 online setting) give the method some independent empirical content, so the score is 6 rather than higher.

Assumptions & free parameters 7 free parameters · 5 assumptions · 1 invented entities

The method's central claim depends on several user-chosen parameters (alpha, gamma, beta, rho, tau, window size, initial concept count) and on domain assumptions about window stationarity and Markov transition structure. The block-diagonal regularizer is an ad hoc heuristic, not a proven constraint, and the 'concept' entity has no independent observable evidence on real datasets.

free parameters (7)
  • alpha = 4
    Trade-off weight in Eq. 2-3; selected from [2,4,6,8,10,20] for 'best performance' on SyD.
  • gamma = 0.8
    Regularization weight for block-diagonal regularizer in Eq. 3; selected from [0.1,0.4,0.8,1,4,10].
  • beta = 60
    Augmented Lagrange relaxation weight in Eq. 10; selected from [5,10,20,40,60,100].
  • rho (concept granularity) = not reported
    Tunable threshold that separates distinct concepts across windows; definition in Appendix A, value not given.
  • tau (eigengap threshold) = not reported
    Threshold in Eq. 8 for estimating the number of concepts k; value not given.
  • window size = dataset-specific (e.g., 78 for SyD, 17 for Stock1)
    Selected via MDL heuristic; for ETT/Traffic/Weather fixed to standard sizes 96/192/336/720.
  • initial concept count per window = 3
    Footnote 4: 'we simply initialize the number of concepts for each window as 3' before estimating k.
assumptions (5)
  • standard math Spectral clustering eigenvalue properties: the number of zero eigenvalues of a graph Laplacian equals the number of connected components.
    Invoked in Sec. 4.1 (Theorem 4.1) and Appendix B to estimate concept count k from the representation matrix.
  • standard math The reproducing kernel Hilbert space and kernel trick are valid for representing nonlinear similarities.
    Used throughout Sec. 3.2 and Sec. 4.1 to justify kernel-induced self-representation.
  • domain assumption Co-evolving time series can be segmented into non-overlapping windows such that each window contains a coherent and stationary concept structure.
    The entire method operates on sliding-window segments; concept drift occurs only between windows. See Sec. 4.2 and Sec. 6.1.
  • domain assumption Concept transitions between consecutive windows follow a first-order Markov process whose probabilities can be estimated from past windows.
    Eqs. 4-5 define transition probabilities from observed transition counts; no evidence is given that real co-evolving series satisfy this stationarity.
  • domain assumption The Gaussian kernel with a global bandwidth captures the true similarity structure of concepts across all datasets.
    Sec. 6.1 and Appendix F fix the Gaussian kernel; the ablation in G.8.4 shows it outperforms other kernels empirically but no principled justification is given.
invented entities (1)
  • Concept (profile pattern)
    purpose: Latent grouping of similar subseries to identify and track drift; central output of the representation matrix.
    A concept is defined by the method itself as a cluster centroid of subseries; on real datasets there is no external ground truth for concept membership, so the entity is only validated indirectly through forecasting performance.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CORAL: Concept Drift Representation Learning for Co-evolving Time-series." pith.science (2026). https://pith.science/paper/BWUMGV25

@misc{pith2026250101480,
  author       = {Pith},
  title        = {Pith review of: CORAL: Concept Drift Representation Learning for Co-evolving Time-series},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BWUMGV25}},
  note         = {Machine review of arXiv:2501.01480}
}
read the original abstract

In the realm of time series analysis, tackling the phenomenon of concept drift poses a significant challenge. Concept drift -- characterized by the evolving statistical properties of time series data, affects the reliability and accuracy of conventional analysis models. This is particularly evident in co-evolving scenarios where interactions among variables are crucial. This paper presents CORAL, a simple yet effective method that models time series as an evolving ecosystem to learn representations of concept drift. CORAL employs a kernel-induced self-representation learning to generate a representation matrix, encapsulating the inherent dynamics of co-evolving time series. This matrix serves as a key tool for identification and adaptation to concept drift by observing its temporal variations. Furthermore, CORAL effectively identifies prevailing patterns and offers insights into emerging trends through pattern evolution analysis. Our empirical evaluation of CORAL across various datasets demonstrates its effectiveness in handling the complexities of concept drift. This approach introduces a novel perspective in the theoretical domain of co-evolving time series analysis, enhancing adaptability and accuracy in the face of dynamic data environments, and can be easily integrated into most deep learning backbones.

Figures

Figures reproduced from arXiv: 2501.01480 by the authors.

Figure 1
Figure 1. Modeling power of CORAL for co-evolving time series: CORAL treats time series as an ecosystem - Œ, and automatically identifies, tracks, and predicts dynamic concepts. (a) Original time series; (b) CORAL-generated representation matrix with distinct concepts (C1-C5) in block diagonal form. The red star marks S1, and purple dashed lines trace S1’s concept drift over time; (c) Identified concepts within the series; (d… view at source ↗
Figure 2
Figure 2. Visualized results on Stock1. As already seen, our method automatically captures typical concepts in a given co-evolving time series ( [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Visualization of N-BEATS and CORAL on ETTh1. ity enables it to effectively capture complex, dynamic in￾teractions within time series data. In contrast, OrbitMap, despite being concept-aware, relies on predefined concepts and struggles with multiple time series, limiting its adapt￾ability. Additionally, [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: Online forecasting results on Stock2 it highly pertinent in the dynamic landscape of financial markets. We conducted tests on the Stock2 dataset [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Partitioning of WS into two sets E and F. The optimal window size is determined by the size corresponding to the border. MSP ELD CCD EQD EOG RDS Stock1 SyD [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]
Figure 6
Figure 6. Figure 6: The best window size (red line) for the eight data sets within the time series data. This involves calculating transition probabilities between concepts and determining the most likely concept transitions, which are then used to forecast future values of the series. F.…
Figure 7
Figure 7. Figure 7: shows the impact of varying α and β on the SyD dataset, while the effect of γ can be found in [PITH_FULL_IMAGE:figures/full_fig_p020_7.png]
Figure 8
Figure 8. Figure 8: (a) Representation matrix with noise. A missing connected block in the lower right corner of the matrix due to noise data. (b-c) Reconstructed representation matrix. This allows us to obtain a noise-reduced representation matrix (b), with the noise captured in (c). G.2…
Figure 9
Figure 9. Figure 9: Discovered concepts for the six real datasets MSP ELD CCD EQD EOG RDS [PITH_FULL_IMAGE:figures/full_fig_p022_9.png]
Figure 10
Figure 10. Figure 10: True (black) and forecasted (blue) values for the six time series, each from a real dataset. G.3. Comprehensive Comparison In this section, we present a detailed comparison of CORAL and Auto-CORAL with thirteen different models. The results are summarized in [PITH_FU…
Figure 11
Figure 11. Figure 11: Results of N-BEATS model. (a) forecasting on three series of Stock1. (b) online forecasting on four series of Stock2. Please also see our results shown in [PITH_FULL_IMAGE:figures/full_fig_p025_11.png]
Figure 12
Figure 12. Figure 12: Box Plot of RMSE Values for Each Model Across Datasets G.6. Distribution of RMSE Values Across all Datasets [PITH_FULL_IMAGE:figures/full_fig_p025_12.png]
Figure 13
Figure 13. Figure 13: Results on random four sequences (pepople1, cars10, 1R2TCR, 2T3RTCR) of Hopkins155 database. The top row shows images from the four sequences with superimposed tracked points. The second row is the heatmap of representation matrices yielded by CORAL. The bottom row is…
Figure 14
Figure 14. Figure 14: The heatmap of representation matrices (binarized) learned on SyD by various SOTA and CORAL. m(m ≪ n) time series, we define two kernel matrices: Kb ∈ R m×m for kernel similarities among the m selected time series, and Ke ∈ R n×m for similarities between the entire da…
Figure 15
Figure 15. Figure 15: Computation time on small/large scale series datasets [PITH_FULL_IMAGE:figures/full_fig_p027_15.png]
Figure 16
Figure 16. Figure 16: The impact of the number of series. G.8. Ablation Study G.8.1. ABLATION STUDY ON THE NUMBER OF SERIES [PITH_FULL_IMAGE:figures/full_fig_p027_16.png]
Figure 17
Figure 17. Figure 17: RMSE across datasets for different regularizations at varying γ clusters can lead to bad representations). Although kernel-based methods are adept at handling nonlinear data, they are helpless in the case of potentially locally manifold structures, e.g., SC, KKM, and …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

67 extracted references · 55 canonical work pages

  1. [1]

    Automatic subspace clustering of high dimensional data

    Agrawal, R., Gehrke, J., Gunopulos, D., and Raghavan, P. Automatic subspace clustering of high dimensional data. Data Mining and Knowledge Discovery, 11: 0 5--33, 2005

    work page 2005

  2. [2]

    Markov-switching garch models in r: The msgarch package

    Ardia, D., Bluteau, K., Boudt, K., Catania, L., and Trottier, D.-A. Markov-switching garch models in r: The msgarch package. Journal of Statistical Software, 91 0 (4), 2019

    work page 2019

  3. [3]

    and Liang, J

    Bai, L. and Liang, J. Sparse subspace clustering with entropy-norm. In International conference on machine learning, pp.\ 561--568. PMLR, 2020

    work page 2020

  4. [4]

    S., and Kassler, A

    Bayram, F., Ahmed, B. S., and Kassler, A. From concept drift to model degradation: An overview on performance-aware drift detectors. Knowledge-Based Systems, 245: 0 108632, 2022

    work page 2022

  5. [5]

    J., and Lucas, A

    Bazzi, M., Blasques, F., Koopman, S. J., and Lucas, A. Time-varying transition probabilities for markov regime switching models. Journal of Time Series Analysis, 38 0 (3): 0 458--478, 2017

    work page 2017

  6. [6]

    and Wang, S

    Bouguessa, M. and Wang, S. Mining projected clusters in high-dimensional spaces. IEEE Transactions on Knowledge and Data Engineering, 21 0 (4): 0 507--522, 2008

    work page 2008

  7. [7]

    Box and jenkins: time series analysis, forecasting and control

    Box, G. Box and jenkins: time series analysis, forecasting and control. In A Very British Affair: Six Britons and the Development of Time Series Analysis During the 20th Century, pp.\ 161--215. Springer, 2013

    work page 2013

  8. [8]

    C., Minku, L

    Cavalcante, R. C., Minku, L. L., and Oliveira, A. L. Fedd: Feature extraction for explicit concept drift detection in time series. In 2016 International Joint Conference on Neural Networks (IJCNN), pp.\ 740--747. IEEE, 2016

    work page 2016

Show all 67 references
  1. [9]

    Self-expressive kernel subspace clustering algorithm for categorical data with embedded feature selection

    Chen, H., Xu, K., Chen, L., and Jiang, Q. Self-expressive kernel subspace clustering algorithm for categorical data with embedded feature selection. Mathematics, 9 0 (14): 0 1680, 2021

    2021

  2. [10]

    and Paschalidis, I

    Chen, R. and Paschalidis, I. Selecting optimal decisions via distributionally robust nearest-neighbor regression. Advances in Neural Information Processing Systems, 32, 2019

    2019

  3. [11]

    Clustering-based cross-sectional regime identification for financial market forecasting

    Chen, R., Sun, M., Xu, K., Patenaude, J.-M., and Wang, S. Clustering-based cross-sectional regime identification for financial market forecasting. In International Conference on Database and Expert Systems Applications, pp.\ 3--16. Springer, 2022 a

    2022

  4. [12]

    Dynamic cross-sectional regime identification for financial market prediction

    Chen, R., Xun, K., Patenaude, J.-M., and Wang, S. Dynamic cross-sectional regime identification for financial market prediction. In 2022 IEEE 46th Annual Computers, Software, and Applications Conference (COMPSAC), pp.\ 295--300. IEEE, 2022 b

    2022

  5. [13]

    Timemae: Self-supervised representations of time series with decoupled masked autoencoders

    Cheng, M., Liu, Q., Liu, Z., Zhang, H., Zhang, R., and Chen, E. Timemae: Self-supervised representations of time series with decoupled masked autoencoders. arXiv preprint arXiv:2303.00320, 2023

    2023

  6. [14]

    Convex optimization & Euclidean distance geometry

    Dattorro, J. Convex optimization & Euclidean distance geometry. Lulu. com, 2010

    2010

  7. [15]

    V., Xue, H., and Salim, F

    Deldari, S., Smith, D. V., Xue, H., and Salim, F. D. Time series change point detection with self-supervised contrastive predictive coding. In Proceedings of the Web Conference 2021, pp.\ 3124--3135, 2021

    2021

  8. [16]

    and Vidal, R

    Elhamifar, E. and Vidal, R. Sparse subspace clustering: Algorithm, theory, and applications. IEEE TPAMI, 35 0 (11): 0 2765--2781, 2013

    2013

  9. [17]

    Dish-ts: a general paradigm for alleviating distribution shift in time series forecasting

    Fan, W., Wang, P., Wang, D., Wang, D., Zhou, Y., and Fu, Y. Dish-ts: a general paradigm for alleviating distribution shift in time series forecasting. In Proceedings of the AAAI Conference on Artificial Intelligence, pp.\ 7522--7529, 2023

    2023

  10. [18]

    T-rep: Representation learning for time series using time-embeddings

    Fraikin, A., Bennetot, A., and Allassonni \`e re, S. T-rep: Representation learning for time series using time-embeddings. arXiv preprint arXiv:2310.04486, 2023

    2023 arXiv

  11. [19]

    Efficient dense subspace clustering

    Ji, P., Salzmann, M., and Li, H. Efficient dense subspace clustering. In IEEE Winter conference on applications of computer vision, pp.\ 461--468. IEEE, 2014

    2014

  12. [20]

    Reversible instance normalization for accurate time-series forecasting against distribution shift

    Kim, T., Kim, J., Tae, Y., Park, C., Choi, J.-H., and Choo, J. Reversible instance normalization for accurate time-series forecasting against distribution shift. In International Conference on Learning Representations, 2021

    2021

  13. [21]

    and Hong, Y

    Li, H. and Hong, Y. Financial volatility forecasting with range-based autoregressive volatility model. Finance Research Letters, 8 0 (2): 0 69--76, 2011

    2011

  14. [22]

    Ddg-da: Data distribution generation for predictable concept drift adaptation

    Li, W., Yang, X., Liu, W., Xia, Y., and Bian, J. Ddg-da: Data distribution generation for predictable concept drift adaptation. In Proceedings of the AAAI Conference on Artificial Intelligence, pp.\ 4092--4100, 2022

    2022

  15. [23]

    S parse TSF : Modeling long-term time series forecasting with *1k* parameters

    Lin, S., Lin, W., Wu, W., Chen, H., and Yang, J. S parse TSF : Modeling long-term time series forecasting with *1k* parameters. In Proceedings of the 41st International Conference on Machine Learning, volume 235 of Proceedings of Machine Learning Research. PMLR, 21--27 Jul 2024

    2024

  16. [24]

    Linearized alternating direction method with adaptive penalty for low-rank representation

    Lin, Z., Liu, R., and Su, Z. Linearized alternating direction method with adaptive penalty for low-rank representation. Advances in neural information processing systems, 24, 2011

    2011

  17. [25]

    Robust recovery of subspace structures by low-rank representation

    Liu, G., Lin, Z., Yan, S., Sun, J., Yu, Y., and Ma, Y. Robust recovery of subspace structures by low-rank representation. IEEE TPAMI, 35 0 (1): 0 171--184, 2012

    2012

  18. [26]

    and Chen, S

    Liu, J. and Chen, S. Timesurl: Self-supervised contrastive learning for universal time series representation learning. In Proceedings of the AAAI Conference on Artificial Intelligence, pp.\ 13918--13926, 2024

    2024

  19. [27]

    Anomaly and change point detection for time series with concept drift

    Liu, J., Yang, D., Zhang, K., Gao, H., and Li, J. Anomaly and change point detection for time series with concept drift. World Wide Web, 26 0 (5): 0 3229--3252, 2023

    2023

  20. [28]

    Subspace clustering by block diagonal representation

    Lu, C., Feng, J., Lin, Z., Mei, T., and Yan, S. Subspace clustering by block diagonal representation. IEEE TPAMI, 41 0 (2): 0 487--501, 2018 a

    2018

  21. [29]

    Learning under concept drift: A review

    Lu, J., Liu, A., Dong, F., Gu, F., Gama, J., and Zhang, G. Learning under concept drift: A review. IEEE transactions on knowledge and data engineering, 31 0 (12): 0 2346--2363, 2018 b

    2018

  22. [30]

    Regions of reliability in the evaluation of multivariate probabilistic forecasts

    Marcotte, \'E ., Zantedeschi, V., Drouin, A., and Chapados, N. Regions of reliability in the evaluation of multivariate probabilistic forecasts. arXiv preprint arXiv:2304.09836, 2023

    2023 arXiv

  23. [31]

    and Sakurai, Y

    Matsubara, Y. and Sakurai, Y. Regime shifts in streams: Real-time forecasting of co-evolving time sequences. In ACM SIGKDD, pp.\ 1045--1054, 2016

    2016

  24. [32]

    and Sakurai, Y

    Matsubara, Y. and Sakurai, Y. Dynamic modeling and forecasting of time-evolving data streams. In ACM SIGKDD, pp.\ 458--468, 2019

    2019

  25. [33]

    and Kajino, H

    Miyaguchi, K. and Kajino, H. Cogra: Concept-drift-aware stochastic gradient descent for time-series forecasting. In Proceedings of the AAAI Conference on Artificial Intelligence, pp.\ 4594--4601, 2019

    2019

  26. [34]

    On spectral clustering: Analysis and an algorithm

    Ng, A., Jordan, M., and Weiss, Y. On spectral clustering: Analysis and an algorithm. Advances in neural information processing systems, 14, 2001

    2001

  27. [35]

    Clustering and projected clustering with adaptive neighbors

    Nie, F., Wang, X., and Huang, H. Clustering and projected clustering with adaptive neighbors. In Proceedings of ACM SIGKDD, pp.\ 977--986, 2014

    2014

  28. [36]

    N., Carpov, D., Chapados, N., and Bengio, Y

    Oreshkin, B. N., Carpov, D., Chapados, N., and Bengio, Y. N-beats: Neural basis expansion analysis for interpretable time series forecasting. arXiv preprint arXiv:1905.10437, 2019

    1905 arXiv

  29. [37]

    Pham, Q., Liu, C., Sahoo, D., and Hoi, S. C. Learning fast and slow for online time series forecasting. arXiv preprint arXiv:2202.11672, 2022

    2022 arXiv

  30. [38]

    Knowledge-maximized ensemble algorithm for different types of concept drift

    Ren, S., Liao, B., Zhu, W., and Li, K. Knowledge-maximized ensemble algorithm for different types of concept drift. Information Sciences, 430: 0 261--281, 2018

    2018

  31. [39]

    Stochastic complexity in statistical inquiry, volume 15

    Rissanen, J. Stochastic complexity in statistical inquiry, volume 15. World scientific, 1998

    1998

  32. [40]

    Stewart, G. W. Matrix perturbation theory. 1990

    1990

  33. [41]

    and Hinton, G

    Van der Maaten, L. and Hinton, G. Visualizing data using t-sne. Journal of machine learning research, 9 0 (11), 2008

    2008

  34. [42]

    A tutorial on spectral clustering

    Von Luxburg, U. A tutorial on spectral clustering. Statistics and computing, 17 0 (4): 0 395--416, 2007

    2007

  35. [43]

    Card: Channel aligned robust blend transformer for time series forecasting

    Wang, X., Zhou, T., Wen, Q., Gao, J., Ding, B., and Jin, R. Card: Channel aligned robust blend transformer for time series forecasting. In The Twelfth International Conference on Learning Representations, 2023

    2023

  36. [44]

    I., Hyde, R., Cao, H., Nguyen, H

    Webb, G. I., Hyde, R., Cao, H., Nguyen, H. L., and Petitjean, F. Characterizing concept drift. Data Mining and Knowledge Discovery, 30 0 (4): 0 964--994, 2016

    2016

  37. [45]

    Onenet: Enhancing time series forecasting models under concept drift by online ensembling

    Wen, Q., Chen, W., Sun, L., Zhang, Z., Wang, L., Jin, R., Tan, T., et al. Onenet: Enhancing time series forecasting models under concept drift by online ensembling. Advances in Neural Information Processing Systems, 36, 2024

    2024

  38. [46]

    Etsformer: Exponential smoothing transformers for time-series forecasting

    Woo, G., Liu, C., Sahoo, D., Kumar, A., and Hoi, S. Etsformer: Exponential smoothing transformers for time-series forecasting. arXiv preprint arXiv:2202.01381, 2022

    2022 arXiv

  39. [47]

    Timesnet: Temporal 2d-variation modeling for general time series analysis

    Wu, H., Hu, T., Liu, Y., Zhou, H., Wang, J., and Long, M. Timesnet: Temporal 2d-variation modeling for general time series analysis. In The eleventh international conference on learning representations, 2022

    2022

  40. [48]

    A self-representation model for robust clustering of categorical sequences

    Xu, K., Chen, L., Wang, S., and Wang, B. A self-representation model for robust clustering of categorical sequences. In Web and Big Data: APWeb-WAIM 2018 International Workshops: MWDA, BAH, KGMA, DMMOOC, DS, Macau, China, July 23--25, 2018, Revised Selected Papers 2, pp.\ 13--...

    2018

  41. [49]

    Kernel subspace clustering algorithm for categorical data

    Xu, K., Chen, L., Sun, H., and Wang, B. Kernel subspace clustering algorithm for categorical data. Journal of Software, 31 0 (11): 0 3492--3505, 2020

    2020

  42. [50]

    Data-driven kernel subspace clustering with local manifold preservation

    Xu, K., Chen, L., and Wang, S. Data-driven kernel subspace clustering with local manifold preservation. In 2022 IEEE International Conference on Data Mining Workshops (ICDMW), pp.\ 876--884. IEEE, 2022 a

    2022

  43. [51]

    A multi-view kernel clustering framework for categorical sequences

    Xu, K., Chen, L., and Wang, S. A multi-view kernel clustering framework for categorical sequences. Expert Systems with Applications, 197: 0 116637, 2022 b

    2022

  44. [52]

    Kernel representation learning with dynamic regime discovery for time series forecasting

    Xu, K., Chen, L., Patenaude, J.-M., and Wang, S. Kernel representation learning with dynamic regime discovery for time series forecasting. In Pacific-Asia Conference on Knowledge Discovery and Data Mining, pp.\ 251--263. Springer, 2024 a

    2024

  45. [53]

    Rhine: A regime-switching model with nonlinear representation for discovering and forecasting regimes in financial markets

    Xu, K., Chen, L., Patenaude, J.-M., and Wang, S. Rhine: A regime-switching model with nonlinear representation for discovering and forecasting regimes in financial markets. In Proceedings of the 2024 SIAM International Conference on Data Mining (SDM), pp.\ 526--534. SIAM, 2024 b

    2024

  46. [54]

    Kan4drift: Are kan effective for identifying and tracking concept drift in time series? In NeurIPS Workshop on Time Series in the Age of Large Models, 2024 c

    Xu, K., Chen, L., and Wang, S. Kan4drift: Are kan effective for identifying and tracking concept drift in time series? In NeurIPS Workshop on Time Series in the Age of Large Models, 2024 c

    2024

  47. [55]

    Kolmogorov-arnold networks for time series: Bridging predictive power and interpretability

    Xu, K., Chen, L., and Wang, S. Kolmogorov-arnold networks for time series: Bridging predictive power and interpretability. arXiv preprint arXiv:2406.02496, 2024 d

    2024 arXiv

  48. [56]

    Wormhole: Concept-aware deep representation learning for co-evolving sequences

    Xu, K., Chen, L., and Wang, S. Wormhole: Concept-aware deep representation learning for co-evolving sequences. arXiv preprint arXiv:2409.13857, 2024 e

    2024 arXiv

  49. [57]

    Towards robust nonlinear subspace clustering: A kernel learning approach

    Xu, K., Chen, L., and Wang, S. Towards robust nonlinear subspace clustering: A kernel learning approach. arXiv preprint arXiv:2501.06368, 2025

    2025 arXiv

  50. [58]

    FITS : Modeling time series with \ 10k\ parameters

    Xu, Z., Zeng, A., and Xu, Q. FITS : Modeling time series with \ 10k\ parameters. In The Twelfth International Conference on Learning Representations, 2024 f

    2024

  51. [59]

    and Hong, S

    Yang, L. and Hong, S. Unsupervised time-series representation learning with iterative bilinear temporal-spectral fusion. In International conference on machine learning, pp.\ 25038--25054. PMLR, 2022

    2022

  52. [60]

    Learning to learn the future: Modeling concept drifts in time series prediction

    You, X., Zhang, M., Ding, D., Feng, F., and Huang, Y. Learning to learn the future: Modeling concept drifts in time series prediction. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management, pp.\ 2434--2443, 2021

    2021

  53. [61]

    Online boosting adaptive learning under concept drift for multistream classification

    Yu, E., Lu, J., Zhang, B., and Zhang, G. Online boosting adaptive learning under concept drift for multistream classification. In Proceedings of the AAAI Conference on Artificial Intelligence, pp.\ 16522--16530, 2024

    2024

  54. [62]

    Are transformers effective for time series forecasting? In Proceedings of the AAAI conference on artificial intelligence, pp.\ 11121--11128, 2023

    Zeng, A., Chen, M., Zhang, L., and Xu, Q. Are transformers effective for time series forecasting? In Proceedings of the AAAI conference on artificial intelligence, pp.\ 11121--11128, 2023

    2023

  55. [63]

    Graph structure fusion for multiview clustering

    Zhan, K., Niu, C., Chen, C., Nie, F., Zhang, C., and Yang, Y. Graph structure fusion for multiview clustering. IEEE Transactions on Knowledge and Data Engineering, 31 0 (10): 0 1984--1993, 2018

    1984

  56. [64]

    Addressing concept shift in online time series forecasting: Detect-then-adapt

    Zhang, Y., Chen, W., Zhu, Z., Qin, D., Sun, L., Wang, X., Wen, Q., Zhang, Z., Wang, L., and Jin, R. Addressing concept shift in online time series forecasting: Detect-then-adapt. arXiv preprint arXiv:2403.14949, 2024

    2024 arXiv

  57. [65]

    Informer: Beyond efficient transformer for long sequence time-series forecasting

    Zhou, H., Zhang, S., Peng, J., Zhang, S., Li, J., Xiong, H., and Zhang, W. Informer: Beyond efficient transformer for long sequence time-series forecasting. In Proceedings of the AAAI conference on artificial intelligence, pp.\ 11106--11115, 2021

    2021

  58. [66]

    Fedformer: Frequency enhanced decomposed transformer for long-term series forecasting

    Zhou, T., Ma, Z., Wen, Q., Wang, X., Sun, L., and Jin, R. Fedformer: Frequency enhanced decomposed transformer for long-term series forecasting. In International conference on machine learning, pp.\ 27268--27286. PMLR, 2022

    2022

  59. [67]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.