Pith. sign in

REVIEW 4 major objections 4 minor 78 references

FIC-TSC: Learning Time Series Classification with Fisher Information Constraint

T0 review · 4 major / 4 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 FIC-TSC claims that constraining a time-series classifier's diagonal Fisher information during training steers it toward flatter minima and improves accuracy under distribution shift, and argues this with theory and results on 115…

desk verdict The method is L2 gradient clipping presented as a Fisher-information constraint; the theory collapses, but the extensive empirical study shows real gains for time series classification. read the letter →

arxiv 2505.06114 v1 pith:2U3RZF6D submitted 2025-05-09 cs.LG

Xiwen Chen , Wenhui Zhu , Peijie Qiu , Hao Wang , Huayu Li , Zihan Li , Yalin Wang , Aristeidis Sotiras
show 1 more author
Abolfazl Razi
This is my paper · ORCID
classification cs.LG
keywords timeseriesclassificationFisherinformationflatminimadomainshiftgradientnormalizationgeneralizationUEAUCR
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

FIC-TSC argues that time-series classifiers fail under train/test distribution shift partly because ordinary training can land in sharp minima, and that a cheap constraint on the diagonal Fisher information of the network guides optimization toward flatter minima. The method re-normalizes gradients whenever the estimated Fisher-information norm exceeds a threshold, requiring no second backward pass and no access to target-domain data. The paper supports this with a sharpness bound linking Fisher information to the curvature of the loss, and with experiments on 30 UEA and 85 UCR datasets where the constrained training improves average accuracy over recent baselines. The practical claim is that a small change to the optimizer—one gradient-scaling step—can substitute for explicit normalization or domain adaptation in time-series classification.

What carries the argument

The central object is the diagonal Fisher information matrix $F(\Theta)$, estimated as the elementwise square of the loss gradient, which measures how much the model's predictions change when each parameter moves. The mechanism is a gradient re-normalization: whenever $\|F\|_1\ge\epsilon$, the update uses $\nabla_\Theta L\leftarrow\sqrt{\epsilon/\|F\|_1}\,\nabla_\Theta L$, so a parameter with a large gradient contribution is scaled down. The identity that carries the argument is Lemma 1, which equates the expected Hessian of the negative log-likelihood with the Fisher information matrix at a local minimum; Corollary 1 then bounds $\alpha$-sharpness by $\alpha^2\|F\|_1/(2(1+L(\Theta)))$. This chain lets the paper claim that constraining Fisher information limits curvature of the loss landscape, pushing training toward flat minima, while keeping the computation at $O(n)$ memory and one backward pass per iteration.

What would settle it

On any UEA dataset, log $\|\nabla L(\Theta_t)\|^2$ and $\nabla L(\Theta_t)^\top \nabla L(\Theta_t)'$ for every iteration $t$; whenever the constraint is active and the minimum of the first quantity is neither $0$ nor $\sqrt{\epsilon}$, the two minima will differ, so the equality used to derive the $O(1/T)$ rate fails. A separate empirical check: train with several values of $\epsilon$ and compare measured sharpness against held-out accuracy; a setting with lower sharpness but equal or worse accuracy would undercut the flat-minima mechanism.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that replacing unconstrained gradient descent with a Fisher-information-constrained update is an efficient way to find flat minima in time-series classifiers. At a local minimum the expected Hessian of the negative log-likelihood equals the Fisher information matrix, and the $\alpha$-sharpness is bounded by $\alpha^2\|F\|_1/(2(1+L(\Theta)))$; capping $\|F\|_1$ therefore caps the sharpness that drives the generalization gap. The implemented constraint is the diagonal approximation $F=\operatorname{diag}((\nabla_\Theta L)\circ(\nabla_\Theta L))$, and the update scales the gradient by $\sqrt{\epsilon/\|F\|_1}$ when the norm exceeds $\epsilon$. The paper reports that this training rule raises average accuracy by 2.1% over prior state of the art on the UEA multivariate benchmark and by 0.2% on the UCR univariate datasets with fixed hyperparameters, reduces measured sharpness by about 40%, and halves runtime compared with sharpness-aware minimization.

Load-bearing premise

The convergence-rate proof assumes that the smallest squared gradient norm over the training run equals the smallest inner product between the raw gradient and the constrained gradient, and that equality stops holding as soon as the Fisher-information constraint is active.

Editorial extensions

If this is right

  • FIC-TSC can be added to an existing time-series classifier by inserting one gradient-scaling step, with no target-domain labels and no second backward pass.
  • If the flat-minima mechanism is right, the same constraint should improve accuracy on other time-series benchmarks where train/test distributions differ, not only the 115 datasets tested.
  • The reported 2.1% average accuracy gain over prior methods on UEA and 0.2% on UCR with fixed hyperparameters is the empirical size of the benefit; the full grid-search version reports 79.6% and 87.3%.
  • The measured sharpness reduction of about 40% on the ten analyzed datasets is a directly observable signature that the method changes the loss landscape rather than merely regularizing outputs.
  • Because only the diagonal Fisher information is used, the method scales to networks with many parameters, and the paper points to a richer block-wise Fisher approximation as a natural follow-up.

Reading between the lines

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

  • Beyond the paper: if the benefit comes from a smaller effective step size rather than from flatness, a baseline trained with a proportionally reduced learning rate should reproduce much of the gain; comparing the two would isolate the mechanism.
  • Beyond the paper: since the Fisher-information-to-sharpness link holds for any smooth loss, the same gradient re-scaling should transfer to forecasting or other classification tasks, and a direct test would be to apply FIC to a non-time-series benchmark with known label shift.
  • Beyond the paper: the fixed threshold $\epsilon$ may deserve scheduling; the paper tests only four fixed values, so an annealed or per-layer $\epsilon$ is a natural unexplored direction.
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

4 major / 4 minor

Summary. The paper proposes FIC-TSC, a training framework for time series classification that constrains the diagonal Fisher information of the network through gradient re-normalization. The authors claim that this constraint guides the optimizer toward flat minima, improves generalization under distribution shift, and preserves an O(1/T) convergence rate. The method is evaluated on 30 UEA multivariate and 85 UCR univariate datasets against 14 recent approaches, with additional ablations, case studies on TimesNet and PatchTST, sharpness measurements, and a comparison with SAM.

Significance. If the theoretical claims were correct, the paper would offer a cheap and architecture-agnostic regularizer for time series classification, backed by an unusually broad empirical evaluation. The empirical work is a genuine strength: the ablation against the same backbone shows consistent gains, the case studies on TimesNet and PatchTST support transferability, the Wilcoxon tests are appropriate for the multi-dataset setting, and the runtime comparison with SAM is informative. However, the central theoretical mechanism is invalid as stated: the quantity called the Fisher information diagonal in Eq. (4) is the square of the batch-averaged gradient, not the Fisher diagonal, and the implemented update reduces exactly to L2 gradient clipping. Because the sharpness and flatness arguments all rest on this identification, the paper's main conceptual contribution is unsupported, and the convergence proof contains an additional unjustified equality.

major comments (4)
  1. [Section 4.2, Eq. (4)] The quantity defined as diag(F(Θ)) = ∇ΘL(Θ) ◦ ∇ΘL(Θ) is the square of the batch-averaged gradient, not the diagonal of the Fisher information matrix; the Fisher diagonal is the average of the squared per-sample gradients. At any differentiable local minimum of the training loss, ∇ΘL = 0, so the paper's diag(F) vanishes even though the Hessian and the true Fisher matrix are generically nonzero. Consequently, Lemma 1 and Corollary 1 (Appendix A, Eqs. (10) and (14)) cannot connect the proposed constraint to the α-sharpness bound, and Proposition 1 is unsupported.
  2. [Section 4.2, Eq. (5)] Because ||F(Θ)||₁ = ||∇ΘL(Θ)||² under the definition in Eq. (4), the normalization step becomes ∇ΘL ← min(1, sqrt(ε)/||∇ΘL||) ∇ΘL, which is exactly standard L2-norm gradient clipping with threshold sqrt(ε). The method is therefore a gradient-norm constraint, not a Fisher information constraint. The reported sharpness reduction in Fig. 8 is also partly self-referential, since the sharpness proxy in Eq. (9) is bounded by the same ||F||₁ = ||∇L||² quantity that the update directly shrinks.
  3. [Appendix A, Eq. (28)] The convergence proof assumes min_t ||∇L(Θ_t)||² = min_t ∇L(Θ_t)ᵀ ∇L(Θ_t)' for the constrained gradient. When the constraint is active, ∇L(Θ_t)' = sqrt(ε/||∇L(Θ_t)||²) ∇L(Θ_t), so ∇L(Θ_t)ᵀ ∇L(Θ_t)' = sqrt(ε)||∇L(Θ_t)||, which is not ||∇L(Θ_t)||². The equality in Eq. (28) therefore fails exactly in the regime the paper is about, and the claimed O(1/T) rate does not follow.
  4. [Appendix G.5] The authors state that non-minimum points do not need to be analyzed because the optimizer can reach local minima in TSC tasks. This is an acknowledged limitation, but it is load-bearing for the flatness claim: the constraint is active throughout training, and the sharpness analysis in Corollary 1 applies only at local minima. As written, the theory does not explain why the training dynamics should converge to a flatter minimum, and the empirical flatness measurement inherits the same gap.
minor comments (4)
  1. [Introduction] The introduction says the method is evaluated on "30 UCR multivariate time series classification datasets" and "85 UEA univariate time series classification datasets," while the abstract and Section 5 state UEA multivariate and UCR univariate; please correct the reversed names.
  2. [Section 4.2] The theorem and remark numbering is inconsistent: the text refers to "Theorem 4" and "Remark 5," while the corresponding proof in Appendix A is labeled Theorem 1; please renumber for consistency.
  3. [Figure 8] The y-axis is labeled "Sharpness Reduction (× 100%)" but the plotted values appear to be around 0.2–0.3, while the text reports an average 40% reduction; please clarify the units or scaling.
  4. [Appendix A, Eq. (17)] The step claiming that the maximum of the quadratic form over the Euclidean ball equals α²||∇²L(Θ)||₂/(2(1+L(Θ))) is stated without derivation; please make the spectral-norm argument explicit.

Circularity Check

2 steps flagged · score 6.0 of 10

Sharpness-reduction evidence is self-referential because Eq. (9) measures the same ||F||_1 that the constraint is built to shrink, and the 'Fisher information' update is standard L2 gradient clipping in disguise; held-out accuracy gains remain independent, so the circularity is partial.

  1. self definitional [Sec. 4.2, Eq. (4) and Corollary 1 Eq. (9); Sec. 6, Landscape Analysis, Fig. 8.]
    "Empirically, given a set of data pairs{Di}n i=1, the FIM is estimated as, diag(F(Θ)) =∇ΘL(Θ)◦∇ ΘL(Θ), (4) ... α-sharpness∝ α2∥F∥1 2(1+L(Θ)). (9) ... Specifically, we compute the sharpness (Eq. 9) between our and the baseline models after well training ... which confirms that using our method can obtain an average 40% reduction in the sharpness across all datasets."

    By Eq. (4), ||F||_1 = ||grad L||^2. The proposed training problem (Eq. 6) directly constrains this norm, so the algorithm is engineered to reduce ||F||_1. Corollary 1 then states that the alpha-sharpness upper bound is proportional to this same ||F||_1. Reporting the Fig. 8 '40% reduction in sharpness' as evidence that FIC finds flatter minima is therefore measuring the exact quantity the constraint shrinks; the flat-minima evidence is a by-construction consequence of the definition, not an independent property of the loss landscape.

  2. renaming known result [Sec. 4.2, Eq. (4)-(6); Algorithm 1.]
    "diag(F(Θ)) =∇ΘL(Θ)◦∇ ΘL(Θ), (4) ... If∥F∥1≥ϵ, the gradient of each parameterθi is normalized as follows: ∇ΘL(Θ)← r ϵ ∥F∥1 ∇ΘL(Θ). (5) ... min Θ L(D; Θ) s.t.∥F(Θ)∥1≤ϵ. (6)"

    Substituting Eq. (4) gives ||F||_1 = sum_j (dL/dtheta_j)^2 = ||grad L||^2. The rescaling in Eq. (5) is therefore grad L <- (sqrt(epsilon)/||grad L||) grad L whenever the norm exceeds sqrt(epsilon), which is exactly standard L2-norm gradient clipping with threshold sqrt(epsilon). The 'Fisher Information Constraint' is thus a re-labeling of a known optimization operation, not a Fisher-information mechanism; the claimed bridge from FIM to sharpness is a name imported after the update rule was defined rather than a quantity the update actually uses.

full rationale

FIC-TSC's headline empirical results are not circular: the 30-UEA and 85-UCR accuracies are evaluated on fixed test splits with universal or grid-searched hyperparameters, the baselines are external, and no test-set fitting or load-bearing self-citation is present; self-citations such as TimeMIL are contextual. What is circular is the sharpness-based support for the flat-minima claim. Eq. (4) defines diag(F) as grad L circle grad L, making ||F||_1 = ||grad L||^2; Corollary 1 then bounds alpha-sharpness by this same norm, and the method is explicitly the constraint ||F||_1 <= epsilon. Consequently, the Fig. 8 '40% sharpness reduction' is a measurement of the quantity the algorithm was built to shrink, i.e., self-referential by construction. The update itself is standard L2 gradient clipping re-labeled as a Fisher information constraint, so the proposed mechanism is a renaming rather than a derived Fisher-geometric effect. Independent content remains in the held-out accuracy improvements and the qualitative contours of Fig. 9, so the circularity is partial rather than total. The proof gap at Eq. (28) and the misidentification of the squared batch-averaged gradient as the Fisher diagonal are additional correctness risks, but they are not additional circularity.

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

The central claims rest on a standard Hessian-Fisher identity, the domain assumption that flat minima generalize, and the specific diagonal FIM approximation that makes the method computationally tractable. The last two are not independently established, and the convergence proof contains an unproven equality that is load-bearing for the theoretical contribution.

free parameters (4)
  • epsilon (FIC threshold) = 2 (universal); grid {2, 4, 10, 20} (Full)
    Controls the strength of gradient re-normalization. Chosen by hand for the universal setting and tuned per dataset in the Full setting; directly determines the clipping threshold.
  • mini-batch size = 64 (UEA) / 16 (UCR) for Uni.; grid {16, 32, 64, 128} for Full
    Standard training hyperparameter that affects gradient estimates and is included in the dataset-wise grid search.
  • learning rate = 5e-3
    AdamW learning rate used for all experiments; a standard hyperparameter chosen by hand.
  • weight decay = 1e-4
    AdamW weight decay used for all experiments; standard regularization hyperparameter.
assumptions (4)
  • standard math The expected Hessian of the negative log-likelihood equals the Fisher Information Matrix at a local minimum (Lemma 1).
    Standard identity under regularity conditions (Kay 1993). The proof in Appendix A assumes grad p(D|Theta)=0, which is a stronger condition than needed, but the result itself is standard.
  • domain assumption Flatter minima imply better generalization under distribution shift.
    Invoked throughout (Section 2, Proposition 1, and the interpretation in Section 6). The paper acknowledges in Appendix G.7 that the flatness-generalization link is an open debate, yet treats it as a working premise.
  • ad hoc to paper The diagonal approximation of the FIM, using squared per-parameter gradients, preserves sufficient information for guiding the optimizer toward flat minima.
    Introduced in Remark 2 and Eq. (4). Justified by computational necessity and analogy to EWC and K-FAC, but not independently verified for this specific purpose. This approximation is load-bearing for the theoretical interpretation.
  • ad hoc to paper A suitable epsilon can be chosen so that min_t ||grad L(Theta_t)||^2 equals min_t grad L(Theta_t)^T grad L(Theta_t)' for the clipped gradient (Appendix A, Eq. 28).
    This equality is generally false when the clipping constraint is active. The O(1/T) convergence proof depends on it, so the proof does not establish the claimed rate.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FIC-TSC: Learning Time Series Classification with Fisher Information Constraint." pith.science (2026). https://pith.science/paper/2U3RZF6D

@misc{pith2026250506114,
  author       = {Pith},
  title        = {Pith review of: FIC-TSC: Learning Time Series Classification with Fisher Information Constraint},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2U3RZF6D}},
  note         = {Machine review of arXiv:2505.06114}
}
read the original abstract

Analyzing time series data is crucial to a wide spectrum of applications, including economics, online marketplaces, and human healthcare. In particular, time series classification plays an indispensable role in segmenting different phases in stock markets, predicting customer behavior, and classifying worker actions and engagement levels. These aspects contribute significantly to the advancement of automated decision-making and system optimization in real-world applications. However, there is a large consensus that time series data often suffers from domain shifts between training and test sets, which dramatically degrades the classification performance. Despite the success of (reversible) instance normalization in handling the domain shifts for time series regression tasks, its performance in classification is unsatisfactory. In this paper, we propose \textit{FIC-TSC}, a training framework for time series classification that leverages Fisher information as the constraint. We theoretically and empirically show this is an efficient and effective solution to guide the model converge toward flatter minima, which enhances its generalizability to distribution shifts. We rigorously evaluate our method on 30 UEA multivariate and 85 UCR univariate datasets. Our empirical results demonstrate the superiority of the proposed method over 14 recent state-of-the-art methods.

Figures

Figures reproduced from arXiv: 2505.06114 by the authors.

Figure 1
Figure 1. A conceptual visualization of Flat and Sharp Minima. The Y-axis indicates loss, and the X-axis represents the variables (neural network parameters). Under train (blue) and test (purple) data domains, due to potential distribution shift, the landscapes differ, i.e., with the same network parameter, the loss is often different. A flat minimum can potentially lead to a low test error, while a sharp minimum potentially … view at source ↗
Figure 2
Figure 2. Histograms representing sequences from the two selected classes of exemplary datasets with train/test distribution shift. The dissimilarity matrix illustrates the min-max normalized Wasserstein-1 distance between class distribution from different sets. A lower value implies two distributions are more similar. It is observed that distribution shifts exist between the entire training and testing sets and within the sa… view at source ↗
Figure 3
Figure 3. An illustration of the negative effect of Instance Nor￾malization (IN). Left: The original input, and Right: Input after applying IN. It is observed that IN can reduce the difference of two class distributions. This may be disadvantageous for classifi￾cation. EC FD HW HB JV SCP1 SCP2 SAD UW PS Avg. 0.0 0.2 0.4 0.6 0.8 1.0 Accuracy Baseline +IN [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4 [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 7
Figure 7. Figure 7: Comparison of the classification accuracy between origi￾nal PatchTST and it after applying our proposed method. EC FD HW HB JV SCP1 SCP2 SAD UW PS Avg. 0.0 0.2 0.4 0.6 0.8 S h a r p n e s s R e d u c tio n (× 1 0 0 %) [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: The sharpness reduction by applying our method. landscape caused by our method. Specifically, we compute the sharpness (Eq. 9) between our and the baseline models after well training. We reiterate here that a lower sharpness often implies a better generalization. We sh…
Figure 5
Figure 5. Figure 5: The accuracy improvement of our method over the base￾line (i.e. standard training without using any constraint). We present the average accuracy improvement at the end. EC FD HW HB JV SCP1 SCP2 SAD UW PS Avg. 0.2 0.4 0.6 0.8 1.0 Accuracy 0.28 0.31 0.68 0.70 0.32 0.34 0…
Figure 9
Figure 9. Figure 9: The landscape (error contour lines) of the well-trained network by baseline (Left) and Ours (Right). The contour lines of our method are more spread out around the center compared to the baseline, indicating a flatter region. Comparison with SAM. We also want to highli…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

78 extracted references · 44 canonical work pages

  1. [1]

    and Flammarion, N

    Andriushchenko, M. and Flammarion, N. Towards understanding sharpness-aware minimization. In International Conference on Machine Learning, pp.\ 639--668. PMLR, 2022

    work page 2022

  2. [2]

    The great time series classification bake off: a review and experimental evaluation of recent algorithmic advances

    Bagnall, A., Lines, J., Bostrom, A., Large, J., and Keogh, E. The great time series classification bake off: a review and experimental evaluation of recent algorithmic advances. Data mining and knowledge discovery, 31: 0 606--660, 2017

    work page 2017

  3. [3]

    A., Lines, J., Flynn, M., Large, J., Bostrom, A., Southam, P., and Keogh, E

    Bagnall, A., Dau, H. A., Lines, J., Flynn, M., Large, J., Bostrom, A., Southam, P., and Keogh, E. The uea multivariate time series classification archive, 2018. arXiv preprint arXiv:1811.00075, 2018

    arXiv 2018

  4. [4]

    Berndt, D. J. and Clifford, J. Using dynamic time warping to find patterns in time series. In Proceedings of the 3rd international conference on knowledge discovery and data mining, pp.\ 359--370, 1994

    1994

  5. [5]

    Homm: Higher-order moment matching for unsupervised domain adaptation

    Chen, C., Fu, Z., Chen, Z., Jin, S., Cheng, Z., Jin, X., and Hua, X.-S. Homm: Higher-order moment matching for unsupervised domain adaptation. In Proceedings of the AAAI conference on artificial intelligence, volume 34, pp.\ 3422--3429, 2020

    work page 2020

  6. [6]

    Timemil: Advancing multivariate time series classification via a time-aware multiple instance learning

    Chen, X., Qiu, P., Zhu, W., Li, H., Wang, H., Sotiras, A., Wang, Y., and Razi, A. Timemil: Advancing multivariate time series classification via a time-aware multiple instance learning. arXiv preprint arXiv:2405.03140, 2024

    arXiv 2024

  7. [7]

    The ucr time series classification archive, July 2015

    Chen, Y., Keogh, E., Hu, B., Begum, N., Bagnall, A., Mueen, A., and Batista, G. The ucr time series classification archive, July 2015. www.cs.ucr.edu/ eamonn/time_series_data/

    work page 2015

  8. [8]

    Contreras-Reyes, J. E. and Kharazmi, O. Belief fisher--shannon information plane: Properties, extensions, and applications to time series analysis. Chaos, Solitons & Fractals, 177: 0 114271, 2023

    work page 2023

Show all 78 references
  1. [9]

    Cover, T. M. Elements of information theory. John Wiley & Sons, 1999

    1999

  2. [10]

    Dempster, A., Petitjean, F., and Webb, G. I. Rocket: exceptionally fast and accurate time series classification using random convolutional kernels. Data Mining and Knowledge Discovery, 34 0 (5): 0 1454--1495, 2020

    2020

  3. [11]

    F., and Webb, G

    Dempster, A., Schmidt, D. F., and Webb, G. I. Hydra: Competing convolutional kernels for fast and accurate time series classification. Data Mining and Knowledge Discovery, 37: 0 1779--1805, 2023

    2023

  4. [12]

    Statistical comparisons of classifiers over multiple data sets

    Dem s ar, J. Statistical comparisons of classifiers over multiple data sets. The Journal of Machine learning research, 7: 0 1--30, 2006

    2006

  5. [13]

    Sharp minima can generalize for deep nets

    Dinh, L., Pascanu, R., Bengio, S., and Bengio, Y. Sharp minima can generalize for deep nets. In International Conference on Machine Learning, pp.\ 1019--1028. PMLR, 2017 a

    2017

  6. [14]

    Sharp minima can generalize for deep nets, 2017 b

    Dinh, L., Pascanu, R., Bengio, S., and Bengio, Y. Sharp minima can generalize for deep nets, 2017 b . URL https://arxiv.org/abs/1703.04933

    2017 arXiv

  7. [15]

    and Abonyi, J

    Dobos, L. and Abonyi, J. Fisher information matrix based time-series segmentation of process data. Chemical Engineering Science, 101: 0 99--108, 2013

    2013

  8. [16]

    K., Cutajar, K., Xie, H., Kandola, J., and Twomey, N

    Early, J., Cheung, G. K., Cutajar, K., Xie, H., Kandola, J., and Twomey, N. Inherently interpretable time series classification via multiple instance learning. In The Twelfth International Conference on Learning Representations, 2024. URL https://openreview.net/forum?id=xriGRsoAza

    2024

  9. [17]

    K., Li, X., and Guan, C

    Eldele, E., Ragab, M., Chen, Z., Wu, M., Kwoh, C. K., Li, X., and Guan, C. Time-series representation learning via temporal and contextual contrasting. arXiv preprint arXiv:2106.14112, 2021

    2021 arXiv

  10. [18]

    Tslanet: Rethinking transformers for time series representation learning

    Eldele, E., Ragab, M., Chen, Z., Wu, M., and Li, X. Tslanet: Rethinking transformers for time series representation learning. arXiv preprint arXiv:2404.08472, 2024

    2024 arXiv

  11. [19]

    O., Ravazzi, C., Dabbene, F., Calafiore, G

    Ferreira, G. O., Ravazzi, C., Dabbene, F., Calafiore, G. C., and Fiore, M. Forecasting network traffic: A survey and tutorial with open-source comparative evaluation. IEEE Access, 11: 0 6018--6044, 2023

    2023

  12. [20]

    Sharpness-aware minimization for efficiently improving generalization

    Foret, P., Kleiner, A., Mobahi, H., and Neyshabur, B. Sharpness-aware minimization for efficiently improving generalization. arXiv preprint arXiv:2010.01412, 2020

    2010 arXiv

  13. [21]

    M., Tan, C

    Foumani, N. M., Tan, C. W., Webb, G. I., and Salehi, M. Improving position encoding of transformers for multivariate time series classification. Data Mining and Knowledge Discovery, 38 0 (1): 0 22--48, 2024

    2024

  14. [22]

    Samformer: Unlocking the potential of transformers in time series forecasting with sharpness-aware minimization and channel-wise attention

    Ilbert, R., Odonnat, A., Feofanov, V., Virmaux, A., Paolo, G., Palpanas, T., and Redko, I. Samformer: Unlocking the potential of transformers in time series forecasting with sharpness-aware minimization and channel-wise attention. In Forty-first International Conference on Mac...

    2024

  15. [23]

    and Szegedy, C

    Ioffe, S. and Szegedy, C. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In International conference on machine learning, pp.\ 448--456. pmlr, 2015

    2015

  16. [24]

    F., Weber, J., Webb, G

    Ismail Fawaz, H., Lucas, B., Forestier, G., Pelletier, C., Schmidt, D. F., Weber, J., Webb, G. I., Idoumghar, L., Muller, P.-A., and Petitjean, F. Inceptiontime: Finding alexnet for time series classification. Data Mining and Knowledge Discovery, 34 0 (6): 0 1936--1962, 2020

    1936

  17. [25]

    Fedfisher: Leveraging fisher information for one-shot federated learning

    Jhunjhunwala, D., Wang, S., and Joshi, G. Fedfisher: Leveraging fisher information for one-shot federated learning. In International Conference on Artificial Intelligence and Statistics, pp.\ 1612--1620. PMLR, 2024

    2024

  18. [26]

    Fantastic generalization measures and where to find them

    Jiang, Y., Neyshabur, B., Mobahi, H., Krishnan, D., and Bengio, S. Fantastic generalization measures and where to find them. arXiv preprint arXiv:1912.02178, 2019

    1912 arXiv

  19. [27]

    Domain adaptation for time series forecasting via attention sharing

    Jin, X., Park, Y., Maddix, D., Wang, H., and Wang, Y. Domain adaptation for time series forecasting via attention sharing. In International Conference on Machine Learning, pp.\ 10280--10297. PMLR, 2022

    2022

  20. [28]

    Multivariate lstm-fcns for time series classification

    Karim, F., Majumdar, S., Darabi, H., and Harford, S. Multivariate lstm-fcns for time series classification. Neural networks, 116: 0 237--245, 2019

    2019

  21. [29]

    Kay, S. M. Statistical signal processing: estimation theory. Prentice Hall, 1: 0 Chapter--3, 1993

    1993

  22. [30]

    H., Tuk, B., Kamphuisen, H

    Kemp, B., Zwinderman, A. H., Tuk, B., Kamphuisen, H. A., and Oberye, J. J. Analysis of a sleep-dependent neuronal feedback loop: the slow-wave microcontinuity of the eeg. IEEE Transactions on Biomedical Engineering, 47 0 (9): 0 1185--1194, 2000

    2000

  23. [31]

    S., Mudigere, D., Nocedal, J., Smelyanskiy, M., and Tang, P

    Keskar, N. S., Mudigere, D., Nocedal, J., Smelyanskiy, M., and Tang, P. T. P. On large-batch training for deep learning: Generalization gap and sharp minima. arXiv preprint arXiv:1609.04836, 2016

    2016 arXiv

  24. [32]

    X., and Hospedales, T

    Kim, M., Li, D., Hu, S. X., and Hospedales, T. Fisher sam: Information geometry and sharpness aware minimisation. In International Conference on Machine Learning, pp.\ 11148--11161. PMLR, 2022 a

    2022

  25. [33]

    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, 2022 b . URL https://openreview.net/forum?id=cGDAkQo1C0p

    2022

  26. [34]

    A., Milan, K., Quan, J., Ramalho, T., Grabska-Barwinska, A., et al

    Kirkpatrick, J., Pascanu, R., Rabinowitz, N., Veness, J., Desjardins, G., Rusu, A. A., Milan, K., Quan, J., Ramalho, T., Grabska-Barwinska, A., et al. Overcoming catastrophic forgetting in neural networks. Proceedings of the national academy of sciences, 114 0 (13): 0 3521--3526, 2017

    2017

  27. [35]

    Overcoming catastrophic forgetting by incremental moment matching

    Lee, S.-W., Kim, J.-H., Jun, J., Ha, J.-W., and Zhang, B.-T. Overcoming catastrophic forgetting by incremental moment matching. Advances in neural information processing systems, 30, 2017

    2017

  28. [36]

    S., Chun, K.-P., and Wong, G

    Li, G., Choi, B., Xu, J., Bhowmick, S. S., Chun, K.-P., and Wong, G. L.-H. Shapenet: A shapelet-neural network approach for multivariate time series classification. In Proceedings of the AAAI conference on artificial intelligence, volume 35, pp.\ 8375--8383, 2021

    2021

  29. [37]

    S., Chen, X., Yuan, G., and Li, A

    Li, H., Carreon-Rascon, A. S., Chen, X., Yuan, G., and Li, A. Mts-lof: medical time-series representation learning via occlusion-invariant features. IEEE Journal of Biomedical and Health Informatics, 2024

    2024

  30. [38]

    Adaptive batch normalization for practical domain adaptation

    Li, Y., Wang, N., Shi, J., Hou, X., and Liu, J. Adaptive batch normalization for practical domain adaptation. Pattern Recognition, 80: 0 109--117, 2018. ISSN 0031-3203. doi:https://doi.org/10.1016/j.patcog.2018.03.005. URL https://www.sciencedirect.com/science/article/pii/S003...

    2018 doi

  31. [39]

    Nutime: Numerically multi-scaled embedding for large-scale time-series pretraining

    Lin, C., Wen, X., Cao, W., Huang, C., Bian, J., Lin, S., and Wu, Z. Nutime: Numerically multi-scaled embedding for large-scale time-series pretraining. arXiv preprint arXiv:2310.07402, 2023

    2023 arXiv

  32. [40]

    Todynet: temporal dynamic graph neural network for multivariate time series classification

    Liu, H., Yang, D., Liu, X., Chen, X., Liang, Z., Wang, H., Cui, Y., and Gu, J. Todynet: temporal dynamic graph neural network for multivariate time series classification. Information Sciences, pp.\ 120914, 2024

    2024

  33. [41]

    Adaptive normalization for non-stationary time series forecasting: A temporal slice perspective

    Liu, Z., Cheng, M., Li, Z., Huang, Z., Liu, Q., Xie, Y., and Chen, E. Adaptive normalization for non-stationary time series forecasting: A temporal slice perspective. Advances in Neural Information Processing Systems, 36: 0 14273--14292, 2023

    2023

  34. [42]

    and Hutter, F

    Loshchilov, I. and Hutter, F. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017

    2017 arXiv

  35. [43]

    P., and Wagenmakers, E.-J

    Ly, A., Marsman, M., Verhagen, J., Grasman, R. P., and Wagenmakers, E.-J. A tutorial on fisher information. Journal of Mathematical Psychology, 80: 0 40--55, 2017

    2017

  36. [44]

    and Grosse, R

    Martens, J. and Grosse, R. Optimizing neural networks with kronecker-factored approximate curvature. In International conference on machine learning, pp.\ 2408--2417. PMLR, 2015

    2015

  37. [45]

    Hive-cote 2.0: a new meta ensemble for time series classification

    Middlehurst, M., Large, J., Flynn, M., Lines, J., Bostrom, A., and Bagnall, A. Hive-cote 2.0: a new meta ensemble for time series classification. Machine Learning, 110 0 (11): 0 3211--3243, 2021

    2021

  38. [46]

    D., Giannakeas, N., and Tzallas, A

    Miltiadous, A., Gionanidis, E., Tzimourta, K. D., Giannakeas, N., and Tzallas, A. T. Dice-net: a novel convolution-transformer architecture for alzheimer detection in eeg signals. IEEe Access, 11: 0 71840--71858, 2023 a

    2023

  39. [47]

    D., Afrantou, T., Ioannidis, P., Grigoriadis, N., Tsalikakis, D

    Miltiadous, A., Tzimourta, K. D., Afrantou, T., Ioannidis, P., Grigoriadis, N., Tsalikakis, D. G., Angelidis, P., Tsipouras, M. G., Glavas, E., Giannakeas, N., et al. A dataset of scalp eeg recordings of alzheimer’s disease, frontotemporal dementia and healthy subjects from ro...

    2023

  40. [48]

    Exploring generalization in deep learning

    Neyshabur, B., Bhojanapalli, S., McAllester, D., and Srebro, N. Exploring generalization in deep learning. Advances in neural information processing systems, 30, 2017

    2017

  41. [49]

    H., Sinthong, P., and Kalagnanam, J

    Nie, Y., Nguyen, N. H., Sinthong, P., and Kalagnanam, J. A time series is worth 64 words: Long-term forecasting with transformers. In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=Jbdc0vTOcol

    2023

  42. [50]

    and Mednick, S

    Niknazar, H. and Mednick, S. C. A multi-level interpretable sleep stage scoring system by infusing experts' knowledge into a deep network architecture. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

    2024

  43. [51]

    Pytorch: An imperative style, high-performance deep learning library

    Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al. Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems, 32, 2019

    2019

  44. [52]

    Relative flatness and generalization

    Petzka, H., Kamp, M., Adilova, L., Sminchisescu, C., and Boley, M. Relative flatness and generalization. Advances in neural information processing systems, 34: 0 18420--18432, 2021

    2021

  45. [53]

    Computational optimal transport: With applications to data science

    Peyr \'e , G., Cuturi, M., et al. Computational optimal transport: With applications to data science. Foundations and Trends in Machine Learning , 11 0 (5-6): 0 355--607, 2019

    2019

  46. [54]

    Variational recurrent adversarial deep domain adaptation

    Purushotham, S., Carvalho, W., Nilanon, T., and Liu, Y. Variational recurrent adversarial deep domain adaptation. In International conference on learning representations, 2017

    2017

  47. [55]

    On wasserstein two-sample testing and related families of nonparametric tests

    Ramdas, A., Garc \' a Trillos, N., and Cuturi, M. On wasserstein two-sample testing and related families of nonparametric tests. Entropy, 19 0 (2): 0 47, 2017

    2017

  48. [56]

    P., Flynn, M., Large, J., Middlehurst, M., and Bagnall, A

    Ruiz, A. P., Flynn, M., Large, J., Middlehurst, M., and Bagnall, A. The great multivariate time series classification bake off: a review and experimental evaluation of recent algorithmic advances. Data Mining and Knowledge Discovery, 35 0 (2): 0 401--449, 2021

    2021

  49. [57]

    Multivariate time series classification using dynamic time warping template selection for human activity recognition

    Seto, S., Zhang, W., and Zhou, Y. Multivariate time series classification using dynamic time warping template selection for human activity recognition. In 2015 IEEE symposium series on computational intelligence, pp.\ 1399--1406. IEEE, 2015

    2015

  50. [58]

    B., Gudelek, M

    Sezer, O. B., Gudelek, M. U., and Ozbayoglu, A. M. Financial time series forecasting with deep learning: A systematic literature review: 2005--2019. Applied soft computing, 90: 0 106181, 2020

    2005

  51. [59]

    and Lovallo, M

    Telesca, L. and Lovallo, M. On the performance of fisher information measure and shannon entropy estimators. Physica A: Statistical Mechanics and its Applications, 484: 0 569--576, 2017

    2017

  52. [60]

    Instance normalization: The missing ingredient for fast stylization

    Ulyanov, D., Vedaldi, A., and Lempitsky, V. Instance normalization: The missing ingredient for fast stylization. arXiv preprint arXiv:1607.08022, 2016

    2016 arXiv

  53. [61]

    The two decades brainclinics research archive for insights in neurophysiology (tdbrain) database

    Van Dijk, H., Van Wingen, G., Denys, D., Olbrich, S., Van Ruth, R., and Arns, M. The two decades brainclinics research archive for insights in neurophysiology (tdbrain) database. Scientific data, 9 0 (1): 0 333, 2022

    2022

  54. [62]

    and Robinson, S

    Vrba, J. and Robinson, S. E. Signal processing in magnetoencephalography. Methods, 25 0 (2): 0 249--271, 2001

    2001

  55. [63]

    I., Samek, W., and Schaeffter, T

    Wagner, P., Strodthoff, N., Bousseljot, R.-D., Kreiseler, D., Lunze, F. I., Samek, W., and Schaeffter, T. Ptb-xl, a large publicly available electrocardiography dataset. Scientific data, 7 0 (1): 0 1--15, 2020

    2020

  56. [64]

    and Shang, P

    Wang, Y. and Shang, P. Analysis of shannon-fisher information plane in time series based on information entropy. Chaos: An Interdisciplinary Journal of Nonlinear Science, 28 0 (10), 2018

    2018

  57. [65]

    Medformer: A multi-granularity patching transformer for medical time-series classification

    Wang, Y., Huang, N., Li, T., Yan, Y., and Zhang, X. Medformer: A multi-granularity patching transformer for medical time-series classification. arXiv preprint arXiv:2405.19363, 2024 a

    2024 arXiv

  58. [66]

    How to evaluate your medical time series classification? arXiv preprint arXiv:2410.03057, 2024 b

    Wang, Y., Li, T., Yan, Y., Song, W., and Zhang, X. How to evaluate your medical time series classification? arXiv preprint arXiv:2410.03057, 2024 b

    2024 arXiv

  59. [67]

    Deep time series models: A comprehensive survey and benchmark

    Wang, Y., Wu, H., Dong, J., Liu, Y., Long, M., and Wang, J. Deep time series models: A comprehensive survey and benchmark. arXiv preprint arXiv:2407.13278, 2024 c

    2024 arXiv

  60. [68]

    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, 2023 a . URL https://openreview.net/forum?id=ju_Uqw384Oq

    2023

  61. [69]

    Interpretable weather forecasting for worldwide stations with a unified deep model

    Wu, H., Zhou, H., Long, M., and Wang, J. Interpretable weather forecasting for worldwide stations with a unified deep model. Nature Machine Intelligence, 5 0 (6): 0 602--611, 2023 b

    2023

  62. [70]

    and He, K

    Wu, Y. and He, K. Group normalization, 2018. URL https://arxiv.org/abs/1803.08494

    2018 arXiv

  63. [71]

    and Yang, E

    Yun, J. and Yang, E. Riemannian sam: sharpness-aware minimization on riemannian manifolds. Advances in Neural Information Processing Systems, 36, 2024

    2024

  64. [72]

    A transformer-based framework for multivariate time series representation learning

    Zerveas, G., Jayaraman, S., Patel, D., Bhamidipaty, A., and Eickhoff, C. A transformer-based framework for multivariate time series representation learning. In Proceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining, pp.\ 2114--2124, 2021

    2021

  65. [73]

    Tapnet: Multivariate time series classification with attentional prototypical network

    Zhang, X., Gao, Y., Lin, J., and Lu, C.-T. Tapnet: Multivariate time series classification with attentional prototypical network. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, pp.\ 6845--6852, 2020

    2020

  66. [74]

    and Yan, J

    Zhang, Y. and Yan, J. Crossformer: Transformer utilizing cross-dimension dependency for multivariate time series forecasting. In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=vSVLM2j9eie

    2023

  67. [75]

    and Xu, Z.-Q

    Zhang, Z. and Xu, Z.-Q. J. Implicit regularization of dropout. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

    2024

  68. [76]

    Domain adaptive ensemble learning

    Zhou, K., Yang, Y., Qiao, Y., and Xiang, T. Domain adaptive ensemble learning. IEEE Transactions on Image Processing, 30: 0 8008–8018, 2021. ISSN 1941-0042. doi:10.1109/tip.2021.3112012. URL http://dx.doi.org/10.1109/TIP.2021.3112012

    2021

  69. [77]

    One fits all: Power general time series analysis by pretrained lm

    Zhou, T., Niu, P., Sun, L., Jin, R., et al. One fits all: Power general time series analysis by pretrained lm. Advances in neural information processing systems, 36: 0 43322--43355, 2023

    2023

  70. [78]

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