pith. sign in

arxiv: 2606.17070 · v1 · pith:HYRUMDOSnew · submitted 2026-06-06 · ⚛️ physics.ao-ph · cs.AI· cs.LG

KFTD: Koopman-Fourier Time-Differentiable Network for Continuous Ocean Spatiotemporal Forecasting

Qinghui Chen , Zekai Zhang , Hailong Liu , Jinglin Zhang , Cong Bai This is my paper

Pith reviewed 2026-06-27 19:09 UTC · model grok-4.3

classification ⚛️ physics.ao-ph cs.AIcs.LG
keywords ocean spatiotemporal forecastingKoopman embeddingFourier interpolationcontinuous-time modelingDPP lossresidual networkphysical consistencydiffusion model comparison
0
0 comments X

The pith

KFTD embeds ocean dynamics in Koopman linear space and uses Fourier interpolation to enable continuous-time forecasting without multi-step sampling.

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

The paper presents a two-stage network that first maps nonlinear ocean dynamics into a linear Koopman space and applies Fourier analysis to generate high-fidelity states at any continuous time sub-step. A lightweight residual network then produces the forecast from those states. This structure removes the need for repeated noise sampling used in diffusion models and adds a DPP loss term that incorporates arbitrary physical equations directly during training. If the approach holds, forecasts for variables such as sea surface temperature become both faster and more consistent with governing equations on large ocean datasets.

Core claim

The central claim is that complex nonlinear ocean dynamics can be mapped into Koopman linear space, with Fourier analysis providing continuous-time interpolation at arbitrary sub-steps; a residual network then evolves these states to produce forecasts. The resulting framework eliminates multi-step noise sampling, supports end-to-end PDE constraints through DPP Loss, and delivers a fourfold computational speedup over diffusion models along with average MSE reductions of 5.6 percent across four ocean datasets.

What carries the argument

Koopman linear embedding combined with Fourier interpolation that produces continuous-time intermediate states for a subsequent residual forecast step.

If this is right

  • Forecasts run four times faster than diffusion models because multi-step noise sampling is removed.
  • Mean squared error drops by 5.6 percent on average and up to 12.7 percent for sea surface temperature across tested ocean datasets.
  • Efficiency improves 76.25 percent relative to MCVD while still allowing arbitrary PDE constraints via the DPP loss.
  • The two-stage separation of interpolation and prediction supports scalable modeling of spatiotemporal fields.

Where Pith is reading between the lines

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

  • The continuous-time formulation could be applied to other fluid or atmospheric systems that require forecasts at irregular observation times.
  • Direct PDE incorporation may reduce the need for post-processing corrections that current data-driven ocean models often require.
  • The linear embedding step might allow easier transfer of trained models across different ocean basins without full retraining.

Load-bearing premise

The Koopman linear embedding plus Fourier interpolation accurately captures the essential nonlinear ocean dynamics at arbitrary time sub-steps without large approximation errors.

What would settle it

High-resolution ground-truth simulations at intermediate time steps show that the interpolated states deviate enough to produce forecast errors larger than those of standard discrete-time models on the same data.

Figures

Figures reproduced from arXiv: 2606.17070 by Cong Bai, Hailong Liu, Jinglin Zhang, Qinghui Chen, Zekai Zhang.

Figure 1
Figure 1. Figure 1: Time-differentiable two-stage framework. (a) Initi [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The schematic representation of the KFTD interpolation model (First stage). As an innovative approach, this model [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The schematic representation of the KFTD predic [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: MSE analysis of predictive models over a seven-step horizon across four datasets. Our model consistently exhibits [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparative visualization of the results from spa [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
read the original abstract

Accurate oceanic forecasting is critical for climate monitoring and disaster early warning. However, ocean spatiotemporal forecasting encounters the double challenges of modeling complex dynamical systems and ensuring computational efficiency. We present Koopman Fourier Time-Differentiable (KFTD) Network, a time continuous twostage paradigm that decouples interpolation from prediction to achieve efficient and scalable spatiotemporal modeling. We map complex nonlinear dynamics into the Koopman linear space and exploit Fourier analysis to enable continuous time interpolation at arbitrary sub-steps. A lightweight residual network consumes the high fidelity intermediate states to yield the final forecast. Unlike diffusion models, KFTD eliminates multi step noise sampling and directly evolves the system in continuous time, yielding a 4 computational speedup. We further introduce a DPP Loss that supports arbitrary PDE constraints in an endtoend manner, breaking the physical consistency bottleneck of pure data-driven approaches. Empirical results on four ocean datasets confirm that our continuous time framework reduces MSE by an average of 5.6% (up to 12.7% for SST) and improves efficiency over MCVD by 76.25%.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 2 minor

Summary. The paper presents the Koopman-Fourier Time-Differentiable (KFTD) Network, a two-stage continuous-time model for ocean spatiotemporal forecasting. It maps nonlinear dynamics to a Koopman linear space, uses Fourier analysis for arbitrary-time interpolation, applies a lightweight residual network for the final forecast, and introduces a DPP Loss to enforce arbitrary PDE constraints. The abstract claims a 4x speedup over diffusion models, 5.6% average MSE reduction (up to 12.7% for SST), and 76.25% efficiency gain over MCVD on four ocean datasets.

Significance. If the empirical claims and architectural assumptions hold under detailed scrutiny, the work could provide a computationally efficient alternative to diffusion-based methods for continuous-time ocean modeling while enabling flexible incorporation of physical constraints. The decoupling of interpolation and residual prediction is a potentially useful paradigm, but its value depends on whether the Koopman-Fourier stage maintains fidelity on nonlinear, multi-scale ocean fields.

major comments (3)
  1. [Abstract] Abstract: The headline performance claims (4x speedup, 5.6% MSE reduction, 76.25% efficiency gain) are presented without equations, dataset descriptions, baseline implementations, error bars, ablation studies, or statistical tests. These metrics are load-bearing for the central empirical contribution and cannot be assessed from the given text.
  2. [Abstract] Abstract (two-stage paradigm): The design assumes the Koopman linear embedding plus Fourier interpolation produces high-fidelity continuous states at arbitrary sub-steps, allowing the residual network to achieve the reported accuracy. No derivation, error bounds, or analysis of approximation error on nonlinear advection/turbulence is supplied, which directly affects whether the efficiency and accuracy claims are valid.
  3. [Abstract] Abstract (DPP Loss): The claim that DPP Loss supports arbitrary PDE constraints in an end-to-end manner is central to overcoming the physical-consistency bottleneck, yet no formulation, enforcement mechanism, or verification against known PDE solutions is provided.
minor comments (2)
  1. [Abstract] Abstract: Typo 'twostage' should be 'two-stage'; '4 computational speedup' should read '4x computational speedup'; 'endtoend' should be 'end-to-end'.
  2. [Abstract] Abstract: Dataset names, sizes, and preprocessing details are omitted, hindering reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment point by point below, clarifying where the supporting material appears in the full text and noting where revisions can strengthen the presentation.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The headline performance claims (4x speedup, 5.6% MSE reduction, 76.25% efficiency gain) are presented without equations, dataset descriptions, baseline implementations, error bars, ablation studies, or statistical tests. These metrics are load-bearing for the central empirical contribution and cannot be assessed from the given text.

    Authors: The abstract is a concise summary; the supporting details are provided in the main manuscript. Dataset descriptions appear in Section 4.1, baseline implementations and efficiency calculations (including the 4x speedup derivation) in Sections 3.5 and 4.2, error bars and statistical tests in Tables 2–3 and Section 5.4, and ablation studies in Section 5.3. We will revise the abstract to include explicit pointers to these sections so readers can immediately locate the supporting material. revision: yes

  2. Referee: [Abstract] Abstract (two-stage paradigm): The design assumes the Koopman linear embedding plus Fourier interpolation produces high-fidelity continuous states at arbitrary sub-steps, allowing the residual network to achieve the reported accuracy. No derivation, error bounds, or analysis of approximation error on nonlinear advection/turbulence is supplied, which directly affects whether the efficiency and accuracy claims are valid.

    Authors: Section 3.2 derives the Koopman-Fourier mapping and continuous interpolation step, while Section 3.3 describes the residual network. Appendix B supplies error bounds and numerical analysis of approximation error on advection and turbulence terms. We agree that a more explicit discussion of limitations for strongly nonlinear regimes would improve clarity and will add a dedicated paragraph in Section 3.2 of the revised manuscript. revision: yes

  3. Referee: [Abstract] Abstract (DPP Loss): The claim that DPP Loss supports arbitrary PDE constraints in an end-to-end manner is central to overcoming the physical-consistency bottleneck, yet no formulation, enforcement mechanism, or verification against known PDE solutions is provided.

    Authors: Section 3.4 gives the mathematical formulation of the DPP Loss, explains the end-to-end enforcement mechanism via the residual network, and Section 5.2 presents verification experiments against known PDE solutions on the ocean datasets. If the current exposition is insufficiently clear, we will expand the derivation and add an additional verification example in the revision. revision: partial

Circularity Check

0 steps flagged

No circularity detected; derivation chain not inspectable from provided text

full rationale

The abstract and reader's summary describe a two-stage architecture (Koopman embedding + Fourier interpolation followed by residual prediction) and report empirical metrics, but supply no equations, loss derivations, or self-citation chains. Without explicit mathematical steps that could reduce a claimed prediction to a fitted input or self-referential definition, no load-bearing circularity is identifiable. The performance claims rest on dataset experiments rather than internal redefinitions, rendering the method self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only review; full paper likely introduces network hyperparameters and loss weights as free parameters. The central claim rests on the unverified assumption that Koopman theory applies directly to ocean dynamics.

axioms (1)
  • domain assumption Nonlinear ocean dynamics can be accurately mapped to a linear Koopman space without loss of essential behavior
    Invoked when stating that the method maps complex nonlinear dynamics into Koopman linear space

pith-pipeline@v0.9.1-grok · 5736 in / 1194 out tokens · 19095 ms · 2026-06-27T19:09:27.139238+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

61 extracted references

  1. [1]

    Hassan Arbabi and Igor Mezić. 2017. Study of dynamics in post-transient flows using Koopman mode decomposition.Physical Review Fluids2, 12 (2017), 124402

    2017

  2. [2]

    Steven L Brunton, Bingni W Brunton, Joshua L Proctor, Eurika Kaiser, and J Nathan Kutz. 2017. Chaos as an intermittently forced linear system.Nature communications8, 1 (2017), 19

    2017

  3. [3]

    Qinghui Chen, Lunqian Wang, Zekai Zhang, Xinghua Wang, Weilin Liu, Bo Xia, Hao Ding, Jinglin Zhang, Sen Xu, and Xin Wang. 2025. Dual-path aggregation transformer network for super-resolution with images occlusions and variability. Engineering Applications of Artificial Intelligence139, PartA (2025)

    2025

  4. [4]

    Qinghui Chen, Zekai Zhang, Zaigui Zhang, Kai Zhang, Dagang Li, Wenmin Wang, Jinglin Zhang, and Cong Liu. 2025. Distilled large language model-driven dynamic sparse expert activation mechanism.Applied Soft Computing(2025), 114037

    2025

  5. [5]

    Zhao Chen, Hao Sun, and Wen Xiong. 2024. Forecasting dynamics by an incom- plete equation of motion and an auto-encoder Koopman operator.Mechanical Systems and Signal Processing220 (2024), 111599

    2024

  6. [6]

    Daniel L Codiga. 2011. Unified tidal analysis and prediction using the UTide Matlab functions. (2011)

    2011

  7. [7]

    Santos Júnior, and Eraylson G Silva

    Paulo SG de Mattos Neto, George DC Cavalcanti, Domingos S de O. Santos Júnior, and Eraylson G Silva. 2022. Hybrid systems using residual modeling for sea surface temperature forecasting.Scientific Reports12, 1 (2022), 487

    2022

  8. [8]

    Akshunna S Dogra and William Redman. 2020. Optimizing neural networks via koopman operator theory.Advances in Neural Information Processing Systems33 (2020), 2087–2097

    2020

  9. [9]

    Runyang Feng, Yixing Gao, Tze Ho Elden Tse, Xueqing Ma, and Hyung Jin Chang

  10. [10]

    InProceedings of the IEEE/CVF International Conference on Computer Vision

    DiffPose: SpatioTemporal diffusion model for video-based human pose estimation. InProceedings of the IEEE/CVF International Conference on Computer Vision. 14861–14872

  11. [11]

    Yuan Feng, Chen Li, and Tianying Sun. 2021. The study based on the deep learning for Indian Ocean Dipole (IOD) index predication. InProceedings of the ACM Turing A ward Celebration Conference-China. 23–27

    2021

  12. [12]

    Yuchao Feng, Jianwei Zheng, Mengjie Qin, Cong Bai, and Jinglin Zhang. 2021. 3D octave and 2D vanilla mixed convolutional neural network for hyperspectral KDD ’26, August 09–13, 2026, Jeju Island, Republic of Korea Qinghui Chen et al. image classification with limited samples.Remote Sensing13, 21 (2021), 4407

    2021

  13. [13]

    Ming Gao, Xiangnan He, Leihui Chen, Tingting Liu, Jinglin Zhang, and Aoy- ing Zhou. 2020. Learning vertex representations for bipartite networks.IEEE transactions on knowledge and data engineering34, 1 (2020), 379–393

    2020

  14. [14]

    Yu Gou, Tong Zhang, Jun Liu, Li Wei, and Jun-Hong Cui. 2020. DeepOcean: A general deep learning framework for spatio-temporal ocean sensing data prediction.IEEE access8 (2020), 79192–79202

    2020

  15. [15]

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. 2020. Denoising diffusion probabilistic models.Advances in neural information processing systems33 (2020), 6840–6851

    2020

  16. [16]

    Cheng Huang, Pan Mu, Jinglin Zhang, Sixian Chan, Shiqi Zhang, Hanting Yan, Shengyong Chen, and Cong Bai. 2025. Benchmark dataset and deep learning method for global tropical cyclone forecasting.Nature Communications16, 1 (2025), 5923

    2025

  17. [17]

    Jinah Kim, Taekyung Kim, and Joon-Gyu Ryu. 2023. Multi-source deep data fusion and super-resolution for downscaling sea surface temperature guided by Generative Adversarial Network-based spatiotemporal dependency learning. International Journal of Applied Earth Observation and Geoinformation119 (2023), 103312

    2023

  18. [18]

    Jiaxun Li, Guihua Wang, Huijie Xue, and Huizan Wang. 2019. A simple predictive model for the eddy propagation trajectory in the northern South China Sea. Ocean Science15, 2 (2019), 401–412

    2019

  19. [19]

    Wei Li, Yuanfu Xie, Shiow-Ming Deng, and Qi Wang. 2010. Application of the multigrid method to the two-dimensional Doppler radar radial velocity data assimilation.Journal of Atmospheric and Oceanic Technology27, 2 (2010), 319– 332

    2010

  20. [20]

    Xin Li, Fusheng Wang, Tao Song, Fan Meng, and Xiaofei Zhao. 2023. ASTMEN: an adaptive spatiotemporal and multi-element fusion network for ocean surface currents forecasting.Frontiers in Marine Science10 (2023), 1281387

    2023

  21. [21]

    Yuanjiang Li, Yi Yang, Kai Zhu, and Jinglin Zhang. 2021. Clothing sale forecast- ing by a composite GRU–Prophet model with an attention mechanism.IEEE Transactions on Industrial Informatics17, 12 (2021), 8335–8344

    2021

  22. [22]

    Ting Liu and Wenxiu Zhong. 2024. Predictability of the upper ocean heat con- tent in a Community Earth System Model ensemble prediction system.Acta Oceanologica Sinica43, 1 (2024), 1–10

    2024

  23. [23]

    Yong Liu, Wenfang Lu, Dong Wang, Zhigang Lai, Chao Ying, Xinwen Li, Ying Han, Zhifeng Wang, and Changming Dong. 2024. Spatiotemporal wave forecast with transformer-based network: A case study for the northwestern Pacific Ocean. Ocean Modelling(2024), 102323

    2024

  24. [24]

    Michael Lutter and Jan Peters. 2023. Combining physics and deep learning to learn continuous-time dynamics models.The International Journal of Robotics Research42, 3 (2023), 83–107

    2023

  25. [25]

    Qing Ma, Cong Bai, Jinglin Zhang, Zhi Liu, and Shengyong Chen. 2019. Super- vised learning based discrete hashing for image retrieval.Pattern Recognition92 (2019), 156–164

    2019

  26. [26]

    Alex T Mallen, Henning Lange, and J Nathan Kutz. 2024. Deep probabilistic Koop- man: long-term time-series forecasting under periodic uncertainties.International Journal of Forecasting40, 3 (2024), 859–868

    2024

  27. [27]

    Minmin Miao, Wenjun Hu, Baoguo Xu, Jinglin Zhang, Joel JPC Rodrigues, and Victor Hugo C De Albuquerque. 2021. Automated CCA-MWF algorithm for unsupervised identification and removal of EOG artifacts from EEG.IEEE Journal of Biomedical and Health Informatics26, 8 (2021), 3607–3617

    2021

  28. [28]

    Fearghal O’Donncha, Yihao Hu, Paulito Palmes, Meredith Burke, Ramon Filgueira, and Jon Grant. 2022. A spatio-temporal LSTM model to forecast across multiple temporal and spatial scales.Ecological Informatics69 (2022), 101687

    2022

  29. [29]

    Sebastian Peitz, Hans Harder, Feliks Nüske, Friedrich M Philipp, Manuel Schaller, and Karl Worthmann. 2024. Equivariance and partial observations in Koopman operator theory for partial differential equations.Journal of Computational Dynamics(2024), 0–0

    2024

  30. [30]

    R Pontoh, M Rafi, C Clorinda, A Ena, M Farras, Restu Arisanti, Toni Toharudin, and Farhat Gumelar. 2024. Deep learning approaches to predict sea surface height above geoid in Pekalongan.International Journal of Data and Network Science8, 2 (2024), 743–752

    2024

  31. [31]

    Maziar Raissi, Paris Perdikaris, and George E Karniadakis. 2019. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations.Journal of Computa- tional physics378 (2019), 686–707

    2019

  32. [32]

    Salva Rühling, Bo Zhao, Hailey Joren, and Rose Yu. 2024. Dyffusion: A dynamics- informed diffusion model for spatiotemporal forecasting.Advances in Neural Information Processing Systems36 (2024)

    2024

  33. [33]

    Qi Shao, Wei Li, Guijun Han, Guangchao Hou, Siyuan Liu, Yantian Gong, and Ping Qu. 2021. A deep learning model for forecasting sea surface height anomalies and temperatures in the South China Sea.Journal of Geophysical Research: Oceans 126, 7 (2021), e2021JC017515

    2021

  34. [34]

    Wei Shao, Yufan Kang, Ziyan Peng, Xiao Xiao, Lei Wang, Yuhui Yang, and Flora D Salim. 2024. STEMO: Early Spatio-temporal Forecasting with Multi-Objective Reinforcement Learning. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 2618–2627

    2024

  35. [35]

    Lu Shi, Masih Haseli, Giorgos Mamakoukas, Daniel Bruder, Ian Abraham, Todd Murphey, Jorge Cortes, and Konstantinos Karydis. 2024. Koopman operators in robot learning.arXiv preprint arXiv:2408.04200(2024)

    arXiv 2024

  36. [36]

    Marzuraikah Mohd Stofa, Mohd Asyraf Zulkifley, and Siti Zulaikha Muhammad Zaki. 2020. A deep learning approach to ship detection using satellite imagery. In IOP Conference Series: Earth and Environmental Science, Vol. 540. IOP Publishing, 012049

    2020

  37. [37]

    Yujin Tang, Peijie Dong, Zhenheng Tang, Xiaowen Chu, and Junwei Liang. 2024. Vmrnn: Integrating vision mamba and lstm for efficient and accurate spatiotem- poral forecasting. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 5663–5673

    2024

  38. [38]

    Polina Verezemskaya, Bernard Barnier, Sergey K Gulev, Sergey Gladyshev, Jean-Marc Molines, Vsevolod Gladyshev, Jean-Michel Lellouche, and Alexan- der Gavrikov. 2021. Assessing Eddying (1/12) ocean reanalysis GLORYS12 using the 14-yr instrumental record from 59.5 N section in the Atlantic.Journal of Geophysical Research: Oceans126, 6 (2021), e2020JC016317

    2021

  39. [39]

    Vikram Voleti, Alexia Jolicoeur-Martineau, and Chris Pal. 2022. Mcvd-masked conditional video diffusion for prediction, generation, and interpolation.Advances in neural information processing systems35 (2022), 23371–23385

    2022

  40. [40]

    Zihao Wang, Guiyong Zhang, Tiezhi Sun, Chongbin Shi, and Bo Zhou. 2023. Data-driven methods for low-dimensional representation and state identification for the spatiotemporal structure of cavitation flow fields.Physics of Fluids35, 3 (2023)

    2023

  41. [41]

    Xinrong Wu. 2016. Improving EnKF-based initialization for ENSO prediction using a hybrid adaptive method.Journal of Climate29, 20 (2016), 7365–7381

    2016

  42. [42]

    Zhen Xing, Qijun Feng, Haoran Chen, Qi Dai, Han Hu, Hang Xu, Zuxuan Wu, and Yu-Gang Jiang. 2024. A survey on video diffusion models.Comput. Surveys 57, 2 (2024), 1–42

    2024

  43. [43]

    Lei Xu, Nengcheng Chen, Zeqiang Chen, Chong Zhang, and Hongchu Yu. 2021. Spatiotemporal forecasting in earth system science: Methods, uncertainties, pre- dictability and future directions.Earth-Science Reviews222 (2021), 103828

    2021

  44. [44]

    Ling Yang, Zhilong Zhang, Yang Song, Shenda Hong, Runsheng Xu, Yue Zhao, Wentao Zhang, Bin Cui, and Ming-Hsuan Yang. 2023. Diffusion models: A comprehensive survey of methods and applications.Comput. Surveys56, 4 (2023), 1–39

    2023

  45. [45]

    Yuting Yang, Junyu Dong, Xin Sun, Estanislau Lima, Quanquan Mu, and Xinhua Wang. 2017. A CFCC-LSTM model for sea surface temperature prediction.IEEE Geoscience and Remote Sensing Letters15, 2 (2017), 207–211

    2017

  46. [46]

    Chin-Chia Michael Yeh, Yujie Fan, Xin Dai, Uday Singh Saini, Vivian Lai, Prince Osei Aboagye, Junpeng Wang, Huiyuan Chen, Yan Zheng, Zhongfang Zhuang, et al. 2024. Rpmixer: Shaking up time series forecasting with random pro- jections for large spatial-temporal data. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 3919–3930

    2024

  47. [47]

    Yuan Yuan, Jingtao Ding, Jie Feng, Depeng Jin, and Yong Li. 2024. Unist: A prompt-empowered universal model for urban spatio-temporal prediction. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 4095–4106

    2024

  48. [48]

    Jinglin Zhang, Pu Liu, Feng Zhang, Hironobu Iwabuchi, Antonio Artur de H e Ayres, Victor Hugo C De Albuquerque, et al . 2020. Ensemble meteorological cloud classification meets internet of dependable and controllable things.IEEE Internet of Things Journal8, 5 (2020), 3323–3330

    2020

  49. [49]

    Jinglin Zhang, Jean-Francois Nezan, and Jean-Gabriel Cousin. 2012. Implementa- tion of motion estimation based on heterogeneous parallel computing system with OpenCL. In2012 IEEE 14th International Conference on High Performance Computing and Communication & 2012 IEEE 9th International Conference on Embedded Software and Systems. IEEE, 41–45

    2012

  50. [50]

    Jinglin Zhang, Zekai Zhang, Qinghui Chen, Gang Li, Weiyu Li, Shijiao Ding, Maomao Xiong, Wenhao Zhang, and Shengyong Chen. 2024. Representation learning based on co-evolutionary combined with probability distribution opti- mization for precise defect location.IEEE Transactions on Neural Networks and Learning Systems36, 7 (2024), 11989–12003

    2024

  51. [51]

    Xudong Zhang, Haoyu Wang, Shuo Wang, Yanliang Liu, Weidong Yu, Jing Wang, Qing Xu, and Xiaofeng Li. 2022. Oceanic internal wave amplitude retrieval from satellite images based on a data-driven transfer learning model.Remote Sensing of Environment272 (2022), 112940

    2022

  52. [52]

    Yiheng Zhang, Ziqiang Wang, Meng Huang, Ming Li, Jian Zhang, Shandong Wang, Jinglin Zhang, and Heng Zhang. 2025. S2DBFT: Spectral-spatial dual-branch fusion transformer for hyperspectral image classification.IEEE Transactions on Geoscience and Remote Sensing(2025)

    2025

  53. [53]

    Zekai Zhang, Qinghui Chen, Maomao Xiong, Shijiao Ding, Zhanzhi Su, Xinjie Yao, Yiming Sun, Cong Bai, and Jinglin Zhang. 2025. Zero-shot learning in industrial scenarios: New large-scale benchmark, challenges and baseline. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. 10357–10366

    2025

  54. [54]

    Zekai Zhang, Gang Li, Haijun Zhang, Qinghui Chen, Qunshu Zhang, Jin Wan, MaoMao Xiong, Cong Bai, Dagang Li, Wenyin Zhang, et al. 2026. A Novel Dataset and Lightweight Distillation Baseline for Highlight Transparent Object Detection. International Journal of Computer Vision134, 4 (2026), 157

    2026

  55. [55]

    Zekai Zhang, Jinglin Zhang, Qinghui Chen, Gang Li, Da Chen, Shuainan Jing, He Wang, Dagang Li, Cong Liu, Cong Bai, et al. 2026. Unification of Closed-Open KFTD: Koopman-Fourier Time-Differentiable Network for Continuous Ocean Spatiotemporal Forecasting KDD ’26, August 09–13, 2026, Jeju Island, Republic of Korea Industrial Detection Scenarios: New Large-Sc...

    2026

  56. [56]

    Zekai Zhang, Mingle Zhou, Honglin Wan, Min Li, Gang Li, and Delong Han

  57. [57]

    Engineering Applications of Artificial Intelligence123 (2023), 106390

    IDD-Net: Industrial defect detection method based on Deep-Learning. Engineering Applications of Artificial Intelligence123 (2023), 106390

    2023

  58. [58]

    Gang Zheng, Xiaofeng Li, Rong-Hua Zhang, and Bin Liu. 2020. Purely satellite data–driven deep learning forecast of complicated tropical instability waves. Science advances6, 29 (2020), eaba1482

    2020

  59. [59]

    Zecheng Zhou, Feng Zhang, Haixia Xiao, Fuchang Wang, Xin Hong, Kun Wu, and Jinglin Zhang. 2021. A novel ground-based cloud image segmentation method by using deep transfer learning.IEEE Geoscience and Remote Sensing Letters19 (2021), 1–5

    2021

  60. [60]

    Pengfei Zhu, Zhilin Zhu, Yu Wang, Jinglin Zhang, and Shuai Zhao. 2022. Multi- granularity episodic contrastive learning for few-shot learning.Pattern Recogni- tion131 (2022), 108820

    2022

  61. [61]

    Nabila Zrira, Assia Kamal-Idrissi, Rahma Farssi, and Haris Ahmad Khan. 2024. Time series prediction of sea surface temperature based on BiLSTM model with attention mechanism.Journal of Sea Research198 (2024), 102472

    2024