Pith. sign in

REVIEW 4 major objections 6 minor 3 cited by

TimeFilter: Patch-Specific Spatial-Temporal Graph Filtration for Time Series Forecasting

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 ↗ Review JSON

Pith's one-line read TimeFilter claims that per-patch filtration of a spatial-temporal graph—keeping only the dependencies each time segment needs—outperforms channel clustering and full dependency modeling across 13 benchmarks.

desk verdict A genuinely new patch-specific MoE filtration mechanism for time-series graphs, with a solid but under-specified experimental write-up; fix the missing hyperparameters and the look-back-search protocol and it deserves acceptance. read the letter →

arxiv 2501.13041 v2 pith:HEXNHMX3 submitted 2025-01-22 cs.LG

Yifan Hu , Guibin Zhang , Peiyuan Liu , Disen Lan , Naiqi Li , Dawei Cheng , Tao Dai , Shu-Tao Xia
show 1 more author
Shirui Pan
This is my paper · ORCID
classification cs.LG
keywords timeseriesforecastinggraphneuralnetworksspatial-temporalgraphspatch-specificfiltrationmixtureofexpertschanneldependencymodelingdynamicroutinglong-term
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

TimeFilter is a graph-neural-network framework for multivariate time series forecasting that treats every short segment (patch) of every channel as a node and learns, for each patch, which dependencies to keep. The paper argues that the two dominant strategies—channel-independent models that ignore inter-channel information and channel-dependent models that fuse all of it—are both wrong for real data, and that coarse channel clustering is too blunt because correlations change over time. TimeFilter constructs a spatial-temporal graph, then uses a mixture-of-experts router with dynamic Top-p allocation to select, per patch, a subset of temporal, spatial, and spatial-temporal edges, filtering out the rest. On 13 datasets spanning weather, electricity, traffic, and solar energy, the paper reports consistent state-of-the-art results, including a 4.48% MSE reduction over the second-best long-term baseline and a 13.54% MSE reduction on the PEMS08 short-term benchmark. The central claim is that patch-specific, dynamically filtered dependencies, rather than any fixed dependency policy, are what improve forecasting accuracy.

What carries the argument

The central object is the patch-specific ego-graph filtration: each of the n = C×N spatial-temporal patches becomes the center of an ego-graph whose edges are partitioned into three regions—temporal (same channel, different patches), spatial (different channels, same time), and spatial-temporal (everything else). A sparsely-gated mixture-of-experts router with dynamic expert allocation (Top-p cumulative confidence) selects a subset of the three filters per patch, producing a sparse adjacency matrix that is then fed to a GNN for aggregation and prediction. The k-NN construction of the initial graph (M = k-NN(GeLU(Dist(X_h)), α), with k = ⌊αn⌋) fixes the candidate edge set before filtration, so all later decisions are choices among those candidates.

What would settle it

Run TimeFilter on a dataset where a single known pair of channels carries all predictive signal, with that pair deliberately placed outside the k nearest neighbors at construction time; if the model still forecasts well, the filtration is not doing the work attributed to it. Alternatively, sweep the pruning factor α over a plausible range (for example 0.05 to 0.5) on Electricity or Traffic and re-run the best baseline under the same sweep: if the reported 4.48% gap shrinks or reverses for any α in that range, the claim that patch-specific filtration drives the gain is falsified.

Watch

Extended reading notes

Core claim

In the paper's own terms, TimeFilter establishes that dependency modeling in multivariate forecasting should be done at the patch level with per-patch filtration, not at the channel level with clustering or with a single global graph. The discovery is that decomposing the initial graph into patch-specific ego-graphs, dividing each into temporal, spatial, and spatial-temporal edge regions, and then letting a noisy-gated router assign one or more filters to each patch—so that the number and type of kept edges varies by patch—yields lower MSE and MAE than any fixed strategy across domains. The paper supports this with ablations showing that heuristic filters (Top-K, random-K, region-wise top-K, channel-wise filtering, or no filtering) all underperform the learned dynamic selection, and with a Wilcoxon test against the second-best baseline at the 99% confidence level.

Load-bearing premise

The learned per-patch filters can only keep or discard edges that survive an initial k-nearest-neighbor pruning of the graph, and the pruning factor α is hand-tuned and never reported; if that pruning deletes a dependency the forecast needs, no later stage can recover it.

Editorial extensions

If this is right

  • Patch-level filtration should replace channel-level clustering as the standard refinement step in multivariate forecasting, since the paper shows it beats CCM and DUET under matched settings.
  • Longer look-back horizons become usable without the usual noise penalty, because filtration removes spurious correlations; the paper demonstrates state-of-the-art results under searched horizons up to 720.
  • The same architecture generalizes across domains with very different correlation structures, from 7-channel ETT to 1763-channel Climate, without changing the dependency policy.
  • Short-term traffic forecasting, where spatial-temporal edges carry the signal, benefits most (13.54% MSE reduction on PEMS08), suggesting the method is especially valuable for sensor-network data.
  • Filtration performed per ego-graph is parallelizable, so the accuracy gain does not require a full dense-graph computation; the paper reports the efficiency is competitive with other GNN-, CNN-, and Transformer-based models.

Reading between the lines

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

  • If patch-specific filtration is the right inductive bias, then the categorical expert split into temporal/spatial/spatial-temporal may be less important than the per-patch sparsity level itself; a learned continuous edge mask with the same dynamic budget could be compared to isolate what carries the gain.
  • The k-NN ceiling implies that combining TimeFilter with a learnable or denser graph constructor (rather than a fixed α) would likely raise accuracy further; this is testable but goes beyond the paper's experiments.
  • The same ego-graph filtration idea transfers to other multivariate sequence tasks such as anomaly detection, imputation, and change-point detection, where the question is also which cross-channel correlations to trust at each moment.
  • Because dynamic Top-p allocation lets each patch choose a different number of experts, the architecture effectively implements a data-dependent sparsity schedule; one could test whether an explicit per-patch sparsity penalty replicates the results without the MoE machinery.
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 / 6 minor

Summary. The paper proposes TimeFilter, a multivariate time series forecasting framework with three modules: (i) a Spatial-Temporal Construction module that patches each channel and builds a k-NN graph over the C×N patch tokens, (ii) a Patch-Specific Filtration module that uses a noisy-gating MoE router with dynamic Top-p allocation to retain only the temporal, spatial, or spatial-temporal edges deemed relevant for each patch, and (iii) an Adaptive Graph Learning module that aggregates the filtered neighborhoods and projects to the forecast. The authors evaluate on 9 long-term and 4 short-term benchmarks and report state-of-the-art results, claiming a 4.48% MSE / 2.23% MAE reduction over Leddam at fixed look-back 96 and a 5.34% / 1.40% reduction over DUET/CCM under searched look-back horizons. The paper includes a Wilcoxon test against Leddam, error bars for that comparison, a six-way ablation of the filtering method, and a released code repository.

Significance. If the reported results are reproducible, TimeFilter offers a meaningful step beyond coarse-grained channel clustering: the patch-specific, dynamically routed filtration of spatial-temporal edges is a novel design in this literature, and the evaluation spans 13 datasets with diverse channel counts and frequencies. The paper deserves credit for releasing code, reporting standard deviations and a Wilcoxon test for the main comparison, and including full per-horizon tables in the appendix. The main weaknesses are that the central mechanism is under-specified (the k-NN sparsity factor α is never reported, loss weights λ1 and λ2 are missing, Top-p choices are incomplete) and that the optimal-horizon protocol and the ablation study are not documented tightly enough to attribute the gains to the proposed filtration rather than to favorable hyperparameter choices or evaluation settings. These are fixable in a revision.

major comments (4)
  1. [4.1, Eq. (3); Table 6; Appendix A.3] The k-NN graph is constructed with k = floor(alpha * n), where alpha is described only as a hand-tuned scaling factor, and no value of alpha appears in the main text, Table 6, or Appendix A.3. This is load-bearing because the Patch-Specific Filtration Module (Section 4.2) can only select edges among those that survive this pruning: a too-small alpha permanently removes useful dependencies, while a too-large alpha makes the hard pruning negligible. The w/o Filter ablation in Table 11 retains the same k-NN graph, so it isolates the filtration module only conditional on an arbitrary alpha, not the adequacy of the graph itself. Please report alpha for every dataset and provide a sensitivity analysis, such as a sweep over alpha with and without filtration, to show that the claimed gains are due to patch-specific filtration rather than a favorable sparsity choice.
  2. [5.2, Table 2; Appendix B.2, Table 9] The optimal-horizon protocol is ambiguous. The text says the look-back length is searched from {192, 336, 512, 720}, but it does not state whether each model is evaluated at its own optimal L or at a single L chosen per dataset, for example the L that is optimal for TimeFilter. Table 9 reports results at all four prediction horizons but does not list which input length was selected for each model and dataset. This matters because the headline improvement of 5.34% and 1.40% over DUET and CCM is obtained under this protocol. Please specify the selection criterion, report the chosen L per model and dataset, and, for robustness, also provide results at a common L.
  3. [4.4, Eq. (18); 5.1] The loss weights lambda1 and lambda2 are defined but never reported for any dataset, and Top-p is said to be selected from {0.0, 0.5} but the per-dataset value is not given. These hyperparameters directly control the routing behavior in Eqs. (9)-(11) and the balance between the prediction loss and the auxiliary losses in Eqs. (16)-(17). Without these values, the experiments are not reproducible from the manuscript alone. Please include a complete per-dataset hyperparameter table that reports alpha, lambda1, lambda2, and Top-p, together with the existing entries in Table 6.
  4. [5.3, Table 4; Appendix B.4, Table 11] The ablation study reports single-run differences without error bars, and several differences are small relative to the reported standard deviations. For example, on Weather the MSE difference between TimeFilter and w/o Filter is 0.239 versus 0.244, compared with a reported plus-or-minus 0.006 standard deviation for TimeFilter in Table 7; on Traffic the difference is 0.407 versus 0.414. It is therefore not established that the filtration module, the dynamic allocation, or the auxiliary losses provide a statistically significant benefit. Please report mean plus-or-minus standard deviation over at least three seeds for the key ablation rows and, ideally, include a paired significance test between TimeFilter and the main competitors in the ablation table.
minor comments (6)
  1. [4.1, Eqs. (2)-(3)] Dist(X_h) is defined as a per-head tensor in R^{H x n x n}, but the text does not explain how the H heads are combined into the single n by n adjacency matrix M in Eq. (3); please state whether the heads are averaged, concatenated, or otherwise aggregated.
  2. [4.1, Eq. (6)] The notation E^S in Eq. (6) lacks the subscript i used in Eqs. (4)-(5); please clarify the indexing.
  3. [Table 4] The header contains the typos Catagories and Weahter; these should be Categories and Weather.
  4. [Table 8] In the Electricity row for prediction horizon 336, the MSE value is rendered as 0,175 with a comma instead of a decimal point; please correct it.
  5. [5.2] The phrase we obtain the p-value of 4.66e-10, indicating a significant improvement at the 99% confidence level is imprecise; a p-value measures evidence under a significance level, so it should be rephrased as significant at the 1% significance level or similar, with the test procedure stated.
  6. [1. Introduction] The contribution statement We for the first time advocate for a fine-grained segmentation of dependencies is an absolute novelty claim; consider softening it unless a systematic prior-art discussion is provided.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: TimeFilter is an empirical architecture evaluated on external benchmarks, with learned routing and filtration that are not fitted to the test set. The k-NN density parameter issue is a sensitivity/robustness concern, not a circularity.

full rationale

TimeFilter is an empirical forecasting method: the graph is constructed from the input via Eq. (3) using k-NN on learned projections, then the Patch-Specific Filtration Module learns a router that selects among temporal, spatial, and spatial-temporal edges, and the model is trained end-to-end with a prediction loss (Eq. 18) on standard train/validation/test splits. There is no theoretical result whose derivation is equivalent to its own assumptions. The reported improvements over baselines are measured on held-out test sets of public benchmarks, so the central claim is not forced by construction. Self-citations appear (e.g., Hu et al. 2024, Liu et al. 2024a, Zhang et al. 2024a,b,c, Dai et al. 2024), but they are used for background motivation, baseline inclusion, or design inspiration and are not load-bearing: removing them would not change the derivation or the benchmark results. The reviewer concern about the unreported scaling factor alpha in Eq. (3) is a legitimate robustness and reproducibility issue, but it does not make the method circular: the router and graph are learned from training data, not fitted to the test set, and the alpha choice does not define the prediction target. No fitted parameter is renamed as a prediction, and no uniqueness theorem or prior same-author result is invoked to forbid alternatives. Accordingly, the appropriate circularity score is 0.

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

The central empirical claim rests on benchmark performance; the main assumptions are architectural (graph edges represent useful dependencies, filtration improves generalization) and experimental (baseline comparisons are fair). Several hyperparameters are hand-tuned and unreported.

free parameters (4)
  • α (kNN scaling factor)
    In Eq. 3, k = floor(α · n) keeps the α fraction of neighbors; α is described as hand-tuned but its value is not listed in Table 6 or Appendix A.3.
  • λ1, λ2 (loss scaling factors)
    In Eq. 18 these weight the dynamic and importance losses; no values are reported.
  • Top-p (dynamic allocation threshold)
    Implementation details say Top-p selected from {0.0, 0.5}, but the actual values used per dataset are not reported.
  • Patch length P = per dataset (e.g., 8 to 96)
    Table 6 lists P per dataset; it is a hand-chosen hyperparameter that controls the granularity of filtration.
assumptions (4)
  • domain assumption Dot-product distances between patch embeddings and k-NN selection produce a graph that contains the dependencies relevant for forecasting.
    Used in Eq. 2 and Eq. 3; if relevant edges are absent from the k-NN graph, the filtration stage cannot recover them.
  • domain assumption Filtering edges by learned per-patch expert selection improves forecast accuracy over keeping all edges.
    This is the core motivation of the PSF module (Section 4.2) and is supported only by ablation, not by theory.
  • domain assumption The optimal look-back horizon exists and can be determined by validation search.
    Invoked for Table 2 and Section 5.4, following the scaling law reference (Shi et al., 2024).
  • domain assumption Baselines in Tables 1-3 are evaluated under comparable protocols.
    The SOTA claim depends on fair comparison; the paper does not state whether baseline numbers were rerun under identical conditions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TimeFilter: Patch-Specific Spatial-Temporal Graph Filtration for Time Series Forecasting." pith.science (2026). https://pith.science/paper/HEXNHMX3

@misc{pith2026250113041,
  author       = {Pith},
  title        = {Pith review of: TimeFilter: Patch-Specific Spatial-Temporal Graph Filtration for Time Series Forecasting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HEXNHMX3}},
  note         = {Machine review of arXiv:2501.13041}
}
read the original abstract

Time series forecasting methods generally fall into two main categories: Channel Independent (CI) and Channel Dependent (CD) strategies. While CI overlooks important covariate relationships, CD captures all dependencies without distinction, introducing noise and reducing generalization. Recent advances in Channel Clustering (CC) aim to refine dependency modeling by grouping channels with similar characteristics and applying tailored modeling techniques. However, coarse-grained clustering struggles to capture complex, time-varying interactions effectively. To address these challenges, we propose TimeFilter, a GNN-based framework for adaptive and fine-grained dependency modeling. After constructing the graph from the input sequence, TimeFilter refines the learned spatial-temporal dependencies by filtering out irrelevant correlations while preserving the most critical ones in a patch-specific manner. Extensive experiments on 13 real-world datasets from diverse application domains demonstrate the state-of-the-art performance of TimeFilter. The code is available at https://github.com/TROUBADOUR000/TimeFilter.

Figures

Figures reproduced from arXiv: 2501.13041 by the authors.

Figure 1
Figure 1. Analysis of three channels from the Electricity dataset shows the pros and cons of CI, CD, and CC strategies in different cases. Dynamic Time Warping (DTW) is a metric for measuring the similarity between two sequences, with lower values indicating higher similarity. The CI strategy ignores the highly correlated co￾variate information in the Supporting Case (left). The CD strategy fuses all information, including th… view at source ↗
Figure 2
Figure 2. The dependency map of 4 different strategies. ti is the time step and xi is one channel. (a) CI strategy preserves only the temporal dependencies. (b) CD strategy fuses all dependencies. (c) Patch-wise Filtration finely selects dependencies for each patch. (d) Channel-wise Clustering coarsely models channel dependencies. TimeFilter begins with Spatial-Temporal Construction Module, which segments the input time serie… view at source ↗
Figure 3
Figure 3. The overall structure of TimeFilter, which consists of: (i) Spatial-Temporal Construction is devised to construct the spatial￾temporal graph from the input X; (ii) Patch-Specific Filtration facilitates spatial-temporal dependencies by filtering out irrelevant information for each patch; (iii) Adaptive Graph Learning is leveraged to predict the future Y based on GNN. n = C × N patches containing the enhanced local in… view at source ↗
Figures from the paper (6 more)
Figure 5
Figure 5. Figure 5: Influence of look-back horizon. TimeFilter consistently outperforms other models under different look-back horizons [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Model efficiency comparison under input-96-predict-96 of ETTm2 and Traffic datasets. model configurations and the same batch size. As shown in [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Visualization of predictions from different models on the Electricity dataset. TimeFilter SOFTS MSGNet Leddam iTransformer PatchTST [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 8
Figure 8. Figure 8: Visualization of predictions from different models on the Traffic dataset. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]
Figure 9
Figure 9. Figure 9: Visualization of predictions from different models on the Weather dataset. TimeFilter SOFTS Crossformer Leddam iTransformer PatchTST [PITH_FULL_IMAGE:figures/full_fig_p019_9.png]
Figure 10
Figure 10. Figure 10: Visualization of predictions from different models on the PEMS08 dataset. 19 [PITH_FULL_IMAGE:figures/full_fig_p019_10.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. JAPE: Joint Anomaly Prediction and Intrinsic Explanation in Multivariate Time Series

    cs.LG 2026-08 conditional novelty 6.0 of 10

    JAPE is a framework that forecasts anomalies in multivariate time series from evolving inter-variable dependency graphs, and reuses those graphs to explain which variables drive each alert.

  2. RDIT: Residual-based Diffusion Implicit Models for Probabilistic Time Series Forecasting

    cs.LG 2025-09 conditional novelty 5.0 of 10

    RDIT adds residual diffusion and variance calibration on top of a strong point forecaster, achieving best CRPS on seven of eight datasets and lower PICP distance in most settings.

  3. Dynamic Perturbed Adaptive Method for Infinite Task-Conflicting Time Series

    cs.LG 2025-05 reject novelty 4.0 of 10

    A trunk-branch method for adapting to conflicting time series tasks reports large error reductions on a synthetic benchmark, but the comparison is confounded by unequal adaptation budgets and the theory overclaims rel...

Reference graph

Works this paper leans on

22 extracted references · 11 canonical work pages · cited by 3 Pith papers

  1. [2]

    I., Hyndman, R

    Godahewa, R., Bergmeir, C., Webb, G. I., Hyndman, R. J., and Montero-Manso, P. Monash time series forecasting archive. arXiv preprint arXiv:2105.06643,

    arXiv

  2. [4]

    Kingma, D. P. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980,

    arXiv

  3. [6]

    e layers

    Hyperparameter settings for different datasets. “e layers” denotes the number of graph block. “lr” denotes the learning rate. “d model” and “d ff” denote the model dimension of attention layers and feed-forward layers, respectively. B. Full Results B.1. Error Bars In this paper, we repeat all the experiments three times. Here we report the standard deviat...

    work page 2024

  4. [8]

    ISBN 9798400708985

    Association for Computing Machinery. ISBN 9798400708985. doi: 10.1145/3633598.3633601. Wang, K., Li, G., Wang, S., Zhang, G., Wang, K., You, Y ., Fang, J., Peng, X., Liang, Y ., and Wang, Y . The snowflake hypothesis: Training and powering gnn with one node one receptive field. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Dis- covery and ...

    work page arXiv

  5. [9]

    The input length L is searched from {192, 336, 512, 720} for optimal horizon in the scaling law of TSF (Shi et al., 2024)

    Full results of long-term forecasting. The input length L is searched from {192, 336, 512, 720} for optimal horizon in the scaling law of TSF (Shi et al., 2024). All results are averaged across four different forecasting horizon: T ∈ {96, 192, 336, 720}. The best and second-best results are highlighted in bold and underlined, respectively. Models TimeFilt...

    work page 2024

  6. [10]

    URL https://doi.org/ 10.1145/3637528.3671766

    1145/3637528.3671766. URL https://doi.org/ 10.1145/3637528.3671766. Wang, Y ., Xu, Y ., Yang, J., Wu, M., Li, X., Xie, L., and Chen, Z. Fully-connected spatial-temporal graph for multivariate time-series data. Proceedings of the AAAI Conference on Artificial Intelligence, 2024b. Wu, H., Hu, T., Liu, Y ., Zhou, H., Wang, J., and Long, M. TimesNet: Temporal...

    arXiv

  7. [11]

    Lsr-igru: Stock trend prediction based on long short-term relationships and improved gru

    Zhu, P., Li, Y ., Hu, Y ., Liu, Q., Cheng, D., and Liang, Y . Lsr-igru: Stock trend prediction based on long short-term relationships and improved gru. International Conference on Information and Knowledge Management, 2024a. Zhu, P., Li, Y ., Hu, Y ., Xiang, S., Liu, Q., Cheng, D., and Liang, Y . Mci-gru: Stock prediction model based on multi-head cross-a...

    arXiv

  8. [12]

    encompasses temperature and power load data from electricity transformers in two regions of China, spanning from 2016 to

    2016

Show all 22 references
  1. [13]

    Sourced from the UCL Machine Learning Repository, this dataset covers the period from 2012 to 2014, providing valuable insights into consumer electricity usage patterns

    features hourly electricity consumption records in kilowatt-hours (kWh) for 321 clients. Sourced from the UCL Machine Learning Repository, this dataset covers the period from 2012 to 2014, providing valuable insights into consumer electricity usage patterns. (4) Traffic datase...

    2012

  2. [14]

    This dataset, covering the years 2015 to 2016, offers a detailed snapshot of traffic flow and congestion

    includes data on hourly road occupancy rates, gathered by 862 detectors across the freeways of the San Francisco Bay area. This dataset, covering the years 2015 to 2016, offers a detailed snapshot of traffic flow and congestion. (5) Solar-Energy dataset (Lai et al.,

    2015

  3. [15]

    (6) Climate dataset (Godahewa et al.,

    contains solar power production data recorded every 10 minutes throughout 2006 from 137 photovoltaic (PV) plants in Alabama. (6) Climate dataset (Godahewa et al.,

    2006

  4. [16]

    We use the Adam optimizer (Kingma, 2014)

    and conducted on an NVIDIA A100 40GB GPU. We use the Adam optimizer (Kingma, 2014). The batch size is set to 16 for the Electricity and Traffic datasets, and 32 for all other datasets. Table 6 provides detailed hyperparameter settings for each dataset. Tasks Dataset Length of ...

    2014

  5. [18]

    The hyperparameter search process involved exploring look-back horizons L ∈ {192, 336, 512, 720}

    of TSF. The hyperparameter search process involved exploring look-back horizons L ∈ {192, 336, 512, 720}. In both settings, TimeFilter consistently achieves the best performance, demonstrating its effectiveness and robustness. In particular, under the optimal horizon condition...

    2024

  6. [19]

    (1) Top-K strategy selects the top K edges with the largest weights from the ego graph Gi of each patch

    We carefully design six other filtering methods to validate the effectiveness of TimeFilter. (1) Top-K strategy selects the top K edges with the largest weights from the ego graph Gi of each patch. (2) Random-K randomly selects K edges from the ego graph Gi of each patch. (3) ...

    2024

  7. [20]

    The input sequence length L is set to 96 for all baselines

    Full results of long-term forecasting. The input sequence length L is set to 96 for all baselines. All results are averaged across four different forecasting horizon: T ∈ {96, 192, 336, 720}. The best and second-best results are highlighted in bold and underlined, respectively...

    2024

  8. [22]

    The input sequence length is set to 96 for all baselines

    Full results of short-term forecasting. The input sequence length is set to 96 for all baselines. All results are averaged across four different forecasting horizon: T ∈ {12, 24, 48}. The best and second-best results are highlighted in bold and underlined, respectively. 16 Tim...

    1990

  9. [2018]

    Segrnn: Segment recurrent neural network for long-term time series forecasting

    Lin, S., Lin, W., Wu, W., Zhao, F., Mo, R., and Zhang, H. Segrnn: Segment recurrent neural network for long-term time series forecasting. arXiv preprint arXiv:2308.11200,

    arXiv

  10. [2019]

    S., Sheng, Z., and Yang, B

    Qiu, X., Hu, J., Zhou, L., Wu, X., Du, J., Zhang, B., Guo, C., Zhou, A., Jensen, C. S., Sheng, Z., and Yang, B. Tfb: Towards comprehensive and fair benchmarking of time series forecasting methods. Proc. VLDB Endow., 2024a. Qiu, X., Wu, X., Lin, Y ., Guo, C., Hu, J., and Yang, ...

    arXiv

  11. [2021]

    SOFTS: Efficient multivariate time series forecasting with series- core fusion

    Han, L., Chen, X.-Y ., Ye, H.-J., and Zhan, D.-C. SOFTS: Efficient multivariate time series forecasting with series- core fusion. In Advances in Neural Information Process- ing Systems, 2024a. Han, L., Ye, H.-J., and Zhan, D.-C. The capacity and ro- bustness trade-off: Revisit...

    arXiv

  12. [2022]

    Timebridge: Non-stationarity matters for long-term time series forecasting

    9 TimeFilter Liu, P., Wu, B., Hu, Y ., Li, N., Dai, T., Bao, J., and Xia, S.-T. Timebridge: Non-stationarity matters for long-term time series forecasting. arXiv preprint arXiv:2410.04442, 2024a. Liu, Y ., Hu, T., Zhang, H., Wu, H., Wang, S., Ma, L., and Long, M. iTransformer:...

    arXiv

  13. [2023]

    Graph sparsification via mixture of graphs

    10 TimeFilter Zhang, G., Sun, X., Yue, Y ., Jiang, C., Wang, K., Chen, T., and Pan, S. Graph sparsification via mixture of graphs. arXiv preprint arXiv:2405.14260, 2024a. Zhang, G., Wang, K., Huang, W., Yue, Y ., Wang, Y ., Zimmermann, R., Zhou, A., Cheng, D., Zeng, J., and Li...

    arXiv

  14. [2024]

    Long- term forecasting with tiDE: Time-series dense encoder

    Das, A., Kong, W., Leach, A., Sen, R., and Yu, R. Long- term forecasting with tiDE: Time-series dense encoder. arXiv preprint arXiv:2304.08424,

    arXiv

Pith tools

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