Pith. sign in

REVIEW 4 major objections 5 minor 39 references

Unveiling the Power of Noise Priors: Enhancing Diffusion Models for Mobile Traffic Prediction

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Decomposing diffusion noise into a data-dynamics prior and a residual improves mobile traffic prediction by over 30%, while speeding training and tightening uncertainty.

desk verdict Simple noise-prior plug-in that likely helps, but the headline 30% needs validation-only hyperparameters and error bars before I'd trust it. read the letter →

arxiv 2501.13794 v3 pith:XYVJPH5K submitted 2025-01-23 cs.LG

Zhi Sheng , Daisy Yuan , Jingtao Ding , Yong Li This is my paper · ORCID
classification cs.LG
keywords mobiletrafficpredictiondiffusionmodelsnoisepriorperiodicdynamicslocalspatio-temporalforecastingdenoisingnetworkcellular
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

Mobile traffic from base stations is strongly periodic and locally autocorrelated, and this paper claims that the noise a diffusion model learns to remove carries those same patterns. The proposed NPDiff framework splits the target noise into a prior, computed directly from extracted periodic and lag-1 dynamics, and a residual that the denoising network still predicts. A weighted blend of the two replaces the pure network output during training and sampling. On four real cellular-traffic datasets, across three different denoising backbones and several forecast horizons, this raises accuracy by more than 30% on average, speeds up early training, shrinks prediction uncertainty, and makes forecasts more robust to input noise. If correct, it means diffusion forecasting for traffic-like time series can be improved simply by engineering the noise, without redesigning the denoising network.

What carries the argument

The load-bearing object is the noise prior $\tilde{\epsilon}$: a closed-form expression that produces a reference noise at every diffusion step from the data dynamics $D$, so that the network only has to learn the leftover. $D$ is chosen from two cheap statistics—a small set of dominant FFT components averaged over the weekly period for multi-step tasks, and the lag-1 previous value for one-step tasks. Fusing via $\hat{\epsilon} = \lambda\tilde{\epsilon} + (1-\lambda)\epsilon_\theta$ with $\lambda \approx 0.5$, the prior anchors the denoising trajectory while the residual keeps it responsive to irregular changes. This is what lets the method be bolted onto CSDI, ConvLSTM, and STID without changing their architectures.

What would settle it

Deliberately break the training periodicity in the test set with a public holiday, network outage, or shifted weekday schedule, and compare NPDiff at a high $\lambda$ against the same model at $\lambda = 0$; if the prior-anchored model degrades below the no-prior baseline in such periods, the assumption that $D$ tracks $x^{ta}_0$ has failed.

Watch

Extended reading notes

Core claim

The central claim is that the Gaussian noise estimated during diffusion denoising is not pure randomness but a carrier of the data's own dynamics, and that exploiting this structure improves prediction. Concretely, with $D$ the extracted dynamics (FFT-based weekly periodicity for multi-step forecasting, lag-1 values for one-step forecasting), the target is written $x^{ta}_0 = D + \Delta x$ with small $\Delta x$. Substituting into the diffusion forward equation gives the noise as a computable prior $\tilde{\epsilon} = (x^{ta}_n - \sqrt{\bar\alpha_n} D)/\sqrt{1-\bar\alpha_n}$ plus a residual $\Delta\epsilon$. NPDiff trains the network against the blended estimate $\hat{\epsilon} = \lambda\tilde{\epsilon} + (1-\lambda)\epsilon_\theta$ and uses the same blend in the reverse update. The paper claims this yields the reported gains while slotting into existing denoising architectures unchanged.

Load-bearing premise

The method assumes that the periodic and one-step-lag patterns measured from the training period are a good stand-in for the specific future values being predicted, so that the leftover $\Delta x$ is small; when that fails, the noise prior would inject a strong and possibly harmful bias.

Editorial extensions

If this is right

  • Any existing diffusion-based traffic predictor can adopt the noise prior as a plug-in, so the reported gains are not tied to a specific denoising architecture.
  • Multi-step forecasting benefits most from the periodic prior, while one-step forecasting benefits most from the lag-1 prior, meaning the choice of dynamics should match the task horizon.
  • Because part of the noise is computed rather than learned, the model converges faster early in training, which matters for real-time resource management.
  • Noise-prior predictions are more stable under injected Gaussian perturbation, indicating greater robustness in noisy operational data.
  • Uncertainty estimates tighten substantially, which improves the reliability of probabilistic forecasts for network operators.

Reading between the lines

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

  • The paper does not test this, but the same decomposition should transfer to other strongly seasonal and autocorrelated flow data such as energy load, water demand, or web traffic.
  • Because the optimal mixing weight lands near 0.5, the learned residual and the data-dynamics prior carry comparable weight; a step-dependent or dataset-adaptive $\lambda$ may extract further gains, which the paper leaves unexplored.
  • A useful check the paper leaves open is to compare against a deterministic model that uses the same periodic and lag-1 dynamics directly, to see how much of the gain comes from the diffusion sampling loop rather than from the dynamics themselves.
  • Under distribution shift—holidays, lockdowns, or cell-site changes—the prior extracted from training data can go stale, so an online-updated $D$ or an uncertainty-gated $\lambda$ would be a natural next step.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes NPDiff, a plug-in framework for diffusion-based mobile traffic forecasting. The central idea is to decompose the diffusion noise ϵ into a data-dynamics-derived "noise prior" eϵ and a residual Δϵ, where eϵ is computed from periodic dynamics (FFT-based, Eq. 4-5) for multi-step prediction and from local dynamics (lag-1 values, Eq. 6) for one-step prediction. The final noise estimate is a weighted blend bϵ = λeϵ + (1−λ)ϵθ of the prior and the denoising network output (Eq. 12). The training objective is modified to predict ϵ against bϵ (Eq. 13), and the reverse update replaces ϵθ with bϵ (Eq. 14). Experiments cover four mobile traffic datasets, three diffusion backbones (CSDI, ConvLSTM, STID), multi-step and one-step tasks, and report an average MAE improvement over the unmodified CSDI baseline of about 41.6% in the 12-12 task, supporting the Abstract's claim of "improvement over 30%." The paper also reports faster early convergence and reduced prediction uncertainty with the noise prior.

Significance. If the reported gains are real, the paper makes a useful conceptual contribution: it identifies that the noise being estimated in a diffusion forecaster can be regularized with data-driven structure, and it demonstrates gains across several denoising architectures. The framework is simple, the derivation in Section 5.3 is algebraically coherent, code and data are promised, and the ablation on λ and NK provides some insight into sensitivity. However, the paper's central quantitative claim rests on three issues that need to be addressed before the result can be accepted at face value: the hyperparameters λ and NK appear to be selected on the test split rather than on a clean validation protocol; the empirical tables have no error bars or significance tests; and the CSDI baseline is anomalously weak on some datasets, so a large fraction of the reported improvement may be repairing a misconfigured baseline. The theoretical guarantee for the modified sampling procedure is also not established, and the uncertainty reduction in Figure 8 is plausibly a shrinkage artifact rather than evidence of better probabilistic calibration.

major comments (4)
  1. [Appendix A.2] Appendix A.2 states that for the model's hyperparameters, NK is chosen from {2,3,5,7,N_mean} and λ from {0.3,...,0.7}, and "we report the best performance selected from all combinations of these two parameter settings." The paper never states that this selection is performed on the validation split. If the selection is made on the test set, all numbers in Tables 2, 3, and 5-9 are optimistically biased, and the headline "improvement over 30%" cannot be distinguished from test-set overfitting. The authors must specify explicitly that selection was on validation data, and ideally report the selected λ and NK per dataset per task. If the selection was on test data, the experiments must be redone with validation-only selection.
  2. [Tables 2, 3, 5-9] The evaluation procedure samples 50 times at test time and reports the average, but no standard deviations, confidence intervals, or significance tests are reported anywhere in the tables. Because diffusion sampling is stochastic and the reported differences are often small (e.g., STID+Prior vs STID on MobileSH14 in Table 5, MAE 0.037 vs 0.046; CSDI+Prior vs CSDI on MobileSH14 in Table 2), the reader cannot determine whether the differences are statistically reliable. The paper should report error bars over the 50 samples or over multiple training seeds, and where differences are small, run paired significance tests. The 90% confidence intervals shown qualitatively in Figure 8 are not sufficient and are not reported numerically for the headline tables.
  3. [Table 2] The CSDI baseline is anomalously weak in several settings: on MobileBJ, CSDI achieves MAE 0.596 and RMSE 3.178, while the simple History Average baseline achieves MAE 0.232 and the STID backbone achieves MAE 0.116; even after adding the prior, CSDI+Prior drops to 0.094. Such a large gap indicates that the CSDI implementation is likely undertuned or misconfigured for this task, and the paper's headline improvement of 41.6% in MAE is computed against this weak baseline. The authors should either tune CSDI with the same care as the baselines, include a stronger diffusion baseline (e.g., DiffSTG or a recent time-series diffusion model), or at minimum report the configuration used for CSDI and show that its poor MobileBJ performance is not due to a fixable implementation issue.
  4. [Section 5.3, Eqs. (7)-(14)] The noise prior eϵ in Eq. (10) is obtained by substituting the extracted dynamics D for the target x_0^ta in the rearranged forward noising equation (Eqs. 7-9). This means the reverse update in Eq. (14) is, by construction, biased toward D when λ > 0. The paper does not provide any analysis of what distribution the modified reverse chain samples, nor does it show that the resulting predictive distribution is calibrated. The reduced uncertainty in Figure 8 may therefore simply reflect shrinkage toward a deterministic seasonal/lag-1 prior rather than an accurate probabilistic model. As a concrete test, the authors should compare NPDiff against a simpler post-hoc procedure that blends the CSDI predictive mean with D (or uses D as an additional conditioning feature) using the same λ. If the simple procedure matches NPDiff's gains, the claim that the noise prior specifically, rather than target shrinkage, is responsible for the improvement would not be supported.
minor comments (5)
  1. [Eq. (1)] The noise schedule is written as {β_n}_{N}^{t=1}; the index should be n=1,...,N. This is likely a typo but should be corrected.
  2. [Abstract and Section 1] There are LaTeX artifacts in the text: "decomposes noise intoprior" (missing space) and "prior}" with a stray brace in the Abstract. These should be cleaned before publication.
  3. [Figures 4 and 5] The axis labels in Figures 4 and 5 render as corrupted Unicode strings such as "/uni00000013/uni00000011..." instead of readable text. The figures need to be regenerated.
  4. [References] The Time-LLM reference [Jin et al.] has no year or venue; please complete the citation.
  5. [Section 6.3] The notation for the number of FFT components is inconsistent: the paper uses NK in Section 5.2 and Appendix A.2, but writes "Nk" in the opening of Section 6.3. Please unify.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the noise prior is a legitimate input feature and the derivation is self-contained algebra.

full rationale

The core derivation is not circular. Eq. (8) is the standard inversion of the forward noising equation; Eq. (9)-(10) define the noise prior by substituting the dynamics D for the target x0, and Eq. (12)-(14) blend this prior with the network noise. Circularity would require D to be derived from the target at test time, but D is computed from training-set periodicity (Eq. 4-5) or the observed lag-1 context (Eq. 6), neither of which is the target. With lambda=0 the method reduces to standard CSDI; with lambda=1 it becomes a deterministic dynamics-based reconstruction, so the method is an interpolation between a legitimate classical feature and a learned model. Moreover, the training objective (Eq. 13) minimizes ||eps - (lambda*e_eps + (1-lambda)*eps_theta)||^2, so a sufficiently expressive network can recover the exact noise and the output is not forced to D. Two correctness risks should be flagged, but they are not circularity: (1) Appendix A.2 says "we report the best performance selected from all combinations of these two parameter settings" without stating that selection is on the validation split; if it is on the test split, the reported gains are optimistically biased. (2) The paper omits a proof that the blended-noise reverse update of Eq. (14) samples the correct predictive distribution, and the CSDI baseline appears anomalously weak on some datasets. These are evaluation and rigor concerns, not self-referential derivations.

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

The method introduces no new physical or mathematical entities. Its reported gains rely on the quality of injected dynamics D and on hyperparameters (λ, NK, P) that are fitted per dataset, with λ and NK selected on the test set. The central assumptions are the closeness of D to the target and the validity of the biased reverse update.

free parameters (3)
  • Noise fusion coefficient λ = Chosen from {0.3, 0.4, 0.5, 0.6, 0.7} per dataset and task, optimal around 0.5 (Figure 4)
    Controls the blend of noise prior and model-predicted noise in Eq. (12). Selected on the test set, it directly affects all reported results.
  • Number of FFT components NK = Set to 5 or NMean (components with amplitude above average) per experiment
    Determines the accuracy of the periodic dynamics prior in Eq. (4). The paper selects the best value across experiments, as stated in Appendix A.2.
  • Period length P for periodic dynamics = One week for most datasets, one day for MobileBJ
    Defines the cycle used to average the FFT-reconstructed signal in Eq. (5). Manually chosen per dataset based on data duration.
assumptions (3)
  • domain assumption Mobile traffic target can be approximated as x^ta_0 = D + Δx with small Δx (Eq. 7).
    Load-bearing: the noise prior eϵ is derived by replacing the target with dynamics D. If D is not close to the target, the residual is large and the prior is a poor reference. Cosine similarities around 0.8 (Table 1) support but do not guarantee this for test periods.
  • domain assumption Using the weighted noise estimate bϵ in the DDPM reverse update (Eq. 14) yields valid samples from the predictive distribution.
    No proof is given that substituting a biased noise estimate into the posterior mean preserves the generative process. This is the theoretical gap behind the 'successful reconstruction' claim in the introduction.
  • standard math The standard DDPM noise schedule and Gaussian noise assumption remain valid for the modified training objective (Eq. 13).
    The reweighted objective is a standard regression, but the resulting bϵ is no longer an unbiased noise estimate at sampling time, so the standard convergence guarantees do not directly apply.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Unveiling the Power of Noise Priors: Enhancing Diffusion Models for Mobile Traffic Prediction." pith.science (2026). https://pith.science/paper/XYVJPH5K

@misc{pith2026250113794,
  author       = {Pith},
  title        = {Pith review of: Unveiling the Power of Noise Priors: Enhancing Diffusion Models for Mobile Traffic Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XYVJPH5K}},
  note         = {Machine review of arXiv:2501.13794}
}
read the original abstract

Accurate prediction of mobile traffic, i.e., network traffic from cellular base stations, is crucial for optimizing network performance and supporting urban development. However, the non-stationary nature of mobile traffic, driven by human activity and environmental changes, leads to both regular patterns and abrupt variations. Diffusion models excel in capturing such complex temporal dynamics due to their ability to capture the inherent uncertainties. Most existing approaches prioritize designing novel denoising networks but often neglect the critical role of noise itself, potentially leading to sub-optimal performance. In this paper, we introduce a novel perspective by emphasizing the role of noise in the denoising process. Our analysis reveals that noise fundamentally shapes mobile traffic predictions, exhibiting distinct and consistent patterns. We propose NPDiff, a framework that decomposes noise into prior and residual components, with the prior} derived from data dynamics, enhancing the model's ability to capture both regular and abrupt variations. NPDiff can seamlessly integrate with various diffusion-based prediction models, delivering predictions that are effective, efficient, and robust. Extensive experiments demonstrate that it achieves superior performance with an improvement over 30\%, offering a new perspective on leveraging diffusion models in this domain. We provide code and data at https://github.com/tsinghua-fib-lab/NPDiff.

Figures

Figures reproduced from arXiv: 2501.13794 by the authors.

Figure 1
Figure 1. Comparison of existing diffusion-based prediction frame [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Noise visualization of the MobileNJ dataset. (a) Noise [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Overview of NPDiff. (a) shows the process of extracting data dynamics, and (b) shows the denoising process with noise priors. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Ablation studies of λ on the MobileSH16 dataset. • Video prediction model performs poorly on mobile traffic data. MAU and MIM, show relatively poor performance compared to other models, indicating that the dynamics of video data differ from those of mobile traffic data…
Figure 6
Figure 6. Figure 6: Comparison of training efficiency on the validation set. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Results of noise perturbation on the MobileSH16 dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 9
Figure 9. Figure 9: Detailed explanation of noise prior in the diffusion process. [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]
Figure 10
Figure 10. Figure 10: Prediction visualization comparing the noise prior-enhanced models and two baseline models (ConvLSTM and STID) on the [PITH_FULL_IMAGE:figures/full_fig_p014_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 28 canonical work pages

  1. [1]

    Adaptive graph convolutional recurrent network for traffic forecasting

    [Bai et al., 2020] Lei Bai, Lina Yao, Can Li, Xianzhi Wang, and Can Wang. Adaptive graph convolutional recurrent network for traffic forecasting. Advances in neural infor- mation processing systems, 33:17804–17815,

    2020

  2. [2]

    • ARIMA: The ARIMA model is a frequently applied sta- tistical approach for time series forecasting

    B Baselines • HA: The History Average approach relies on the mean of data from previous time intervals to forecast target values. • ARIMA: The ARIMA model is a frequently applied sta- tistical approach for time series forecasting. This method effectively predicts time series data obtained at consistent time intervals. • PatchTST [Nie et al., 2022 ]: It in...

    work page 2022

  3. [6]

    Pre- serve your own correlation: A noise prior for video dif- fusion models

    [Ge et al., 2023] Songwei Ge, Seungjun Nah, Guilin Liu, Tyler Poon, Andrew Tao, Bryan Catanzaro, David Jacobs, Jia-Bin Huang, Ming-Yu Liu, and Yogesh Balaji. Pre- serve your own correlation: A noise prior for video dif- fusion models. In Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision , pages 22930– 22941,

    work page 2023

  4. [7]

    Denoising diffusion probabilistic models

    [Ho et al., 2020] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems , 33:6840–6851,

    2020

  5. [8]

    Video diffusion models

    [Ho et al., 2022] Jonathan Ho, Tim Salimans, Alexey Grit- senko, William Chan, Mohammad Norouzi, and David J Fleet. Video diffusion models. Advances in Neural Infor- mation Processing Systems, 35:8633–8646,

    work page 2022

  6. [11]

    Attentive crowd flow machines

    [Liu et al., 2018] Lingbo Liu, Ruimao Zhang, Jiefeng Peng, Guanbin Li, Bowen Du, and Liang Lin. Attentive crowd flow machines. In Proceedings of the 26th ACM interna- tional conference on Multimedia, pages 1553–1561,

    work page 2018

  7. [12]

    itransformer: Inverted transformers are effective for time series forecasting

    [Liu et al., 2023] Yong Liu, Tengge Hu, Haoran Zhang, Haixu Wu, Shiyu Wang, Lintao Ma, and Mingsheng Long. itransformer: Inverted transformers are effective for time series forecasting. arXiv preprint arXiv:2310.06625,

    arXiv 2023

  8. [13]

    Improved denoising diffusion proba- bilistic models

    [Nichol and Dhariwal, 2021] Alexander Quinn Nichol and Prafulla Dhariwal. Improved denoising diffusion proba- bilistic models. In International conference on machine learning, pages 8162–8171. PMLR,

    work page 2021

Show all 39 references
  1. [14]

    A time series is worth 64 words: Long-term forecasting with transformers

    [Nie et al., 2022] Yuqi Nie, Nam H Nguyen, Phanwadee Sinthong, and Jayant Kalagnanam. A time series is worth 64 words: Long-term forecasting with transformers. arXiv preprint arXiv:2211.14730,

    2022 arXiv

  2. [15]

    Freenoise: Tuning-free longer video diffusion via noise rescheduling

    [Qiu et al., 2023] Haonan Qiu, Menghan Xia, Yong Zhang, Yingqing He, Xintao Wang, Ying Shan, and Ziwei Liu. Freenoise: Tuning-free longer video diffusion via noise rescheduling. arXiv preprint arXiv:2310.15169,

    2023 arXiv

  3. [16]

    Mo- bile data science and intelligent apps: concepts, ai-based modeling and research directions

    [Sarker et al., 2021] Iqbal H Sarker, Mohammed Moshiul Hoque, Md Kafil Uddin, and Tawfeeq Alsanoosy. Mo- bile data science and intelligent apps: concepts, ai-based modeling and research directions. Mobile Networks and Applications, 26(1):285–303,

    2021

  4. [17]

    Spatial-temporal identity: A simple yet effective baseline for multivariate time series forecast- ing

    [Shao et al., 2022] Zezhi Shao, Zhao Zhang, Fei Wang, Wei Wei, and Yongjun Xu. Spatial-temporal identity: A simple yet effective baseline for multivariate time series forecast- ing. In Proceedings of the 31st ACM International Con- ference on Information & Knowledge Management...

    2022

  5. [18]

    Non- autoregressive conditional diffusion models for time se- ries prediction

    [Shen and Kwok, 2023] Lifeng Shen and James Kwok. Non- autoregressive conditional diffusion models for time se- ries prediction. In International Conference on Machine Learning, pages 31016–31029. PMLR,

    2023

  6. [19]

    Convolutional lstm network: A machine learning approach for precipitation nowcasting

    [Shi et al., 2015] Xingjian Shi, Zhourong Chen, Hao Wang, Dit-Yan Yeung, Wai-Kin Wong, and Wang-chun Woo. Convolutional lstm network: A machine learning approach for precipitation nowcasting. Advances in neural informa- tion processing systems, 28,

    2015

  7. [21]

    Csdi: Conditional score- based diffusion models for probabilistic time series impu- tation

    [Tashiro et al., 2021] Yusuke Tashiro, Jiaming Song, Yang Song, and Stefano Ermon. Csdi: Conditional score- based diffusion models for probabilistic time series impu- tation. Advances in Neural Information Processing Sys- tems, 34:24804–24816,

    2021

  8. [22]

    Predrnn: Recurrent neural networks for predictive learning using spatiotem- poral lstms

    [Wang et al., 2017] Yunbo Wang, Mingsheng Long, Jianmin Wang, Zhifeng Gao, and Philip S Yu. Predrnn: Recurrent neural networks for predictive learning using spatiotem- poral lstms. Advances in neural information processing systems, 30,

    2017

  9. [25]

    A survey on video diffusion models

    [Xing et al., 2023] Zhen Xing, Qijun Feng, Haoran Chen, Qi Dai, Han Hu, Hang Xu, Zuxuan Wu, and Yu-Gang Jiang. A survey on video diffusion models. ACM Com- puting Surveys,

    2023

  10. [27]

    Network traffic overload prediction with tempo- ral graph attention convolutional networks

    [Yu et al., 2022] Qiaohong Yu, Huandong Wang, Tong Li, Depeng Jin, Xing Wang, Lin Zhu, Junlan Feng, and Chao Deng. Network traffic overload prediction with tempo- ral graph attention convolutional networks. In 2022 IEEE International Conference on Communications Workshops (ICC...

    2022

  11. [28]

    Spatio-temporal diffu- sion point processes

    [Yuan et al., 2023] Yuan Yuan, Jingtao Ding, Chenyang Shao, Depeng Jin, and Yong Li. Spatio-temporal diffu- sion point processes. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 3173–3184,

    2023

  12. [29]

    A foundation model for unified urban spatio-temporal flow prediction

    [Yuan et al., 2024b] Yuan Yuan, Jingtao Ding, Chonghua Han, Depeng Jin, and Yong Li. A foundation model for unified urban spatio-temporal flow prediction. arXiv preprint arXiv:2411.12972,

    arXiv

  13. [30]

    Urbandit: A founda- tion model for open-world urban spatio-temporal learning

    [Yuan et al., 2024c] Yuan Yuan, Chonghua Han, Jingtao Ding, Depeng Jin, and Yong Li. Urbandit: A founda- tion model for open-world urban spatio-temporal learning. arXiv preprint arXiv:2411.12164,

  14. [31]

    Deep spatio-temporal residual networks for citywide crowd flows prediction

    [Zhang et al., 2017] Junbo Zhang, Yu Zheng, and Dekang Qi. Deep spatio-temporal residual networks for citywide crowd flows prediction. In Proceedings of the AAAI con- ference on artificial intelligence, volume 31,

    2017

  15. [32]

    Promptst: Prompt- enhanced spatio-temporal multi-attribute prediction

    [Zhang et al., 2023] Zijian Zhang, Xiangyu Zhao, Qidong Liu, Chunxu Zhang, Qian Ma, Wanyu Wang, Hongwei Zhao, Yiqi Wang, and Zitao Liu. Promptst: Prompt- enhanced spatio-temporal multi-attribute prediction. In Proceedings of the 32nd ACM International Conference on Information...

    2023

  16. [33]

    Trip: Temporal residual learning with image noise prior for image-to-video diffusion models

    [Zhang et al., 2024] Zhongwei Zhang, Fuchen Long, Ying- wei Pan, Zhaofan Qiu, Ting Yao, Yang Cao, and Tao Mei. Trip: Temporal residual learning with image noise prior for image-to-video diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- ter...

    2024

  17. [34]

    T-gcn: A temporal graph convolutional network for traffic prediction

    [Zhao et al., 2019] Ling Zhao, Yujiao Song, Chao Zhang, Yu Liu, Pu Wang, Tao Lin, Min Deng, and Haifeng Li. T-gcn: A temporal graph convolutional network for traffic prediction. IEEE transactions on intelligent transportation systems, 21(9):3848–3858,

    2019

  18. [35]

    St-gsp: Spatial-temporal global semantic represen- tation learning for urban flow prediction

    [Zhao et al., 2022] Liang Zhao, Min Gao, and Zongwei Wang. St-gsp: Spatial-temporal global semantic represen- tation learning for urban flow prediction. In Proceedings of the Fifteenth ACM International Conference on Web Search and Data Mining, pages 1443–1451,

    2022

  19. [36]

    Towards generative modeling of urban flow through knowledge-enhanced denoising diffu- sion

    [Zhou et al., 2023] Zhilun Zhou, Jingtao Ding, Yu Liu, De- peng Jin, and Yong Li. Towards generative modeling of urban flow through knowledge-enhanced denoising diffu- sion. In Proceedings of the 31st ACM International Con- ference on Advances in Geographic Information Systems...

    2023

  20. [37]

    Coms2t: A complementary spatiotemporal learning system for data-adaptive model evolution

    [Zhou et al., 2024] Zhengyang Zhou, Qihe Huang, Binwu Wang, Jianpeng Hou, Kuo Yang, Yuxuan Liang, and Yang Wang. Coms2t: A complementary spatiotemporal learning system for data-adaptive model evolution. arXiv preprint arXiv:2403.01738,

    2024 arXiv

  21. [64]

    We use a batch size of 8 and an initial learning rate of 1e-3, which is reduced to 4e-4 after 40 epochs

    During training, we employ a quadratic noise schedule in our diffusion model, starting with a noise level ofβ1 = 0.0001, which increases to a maximum of βN = 0.5 over N = 50 steps. We use a batch size of 8 and an initial learning rate of 1e-3, which is reduced to 4e-4 after 40...

    2021

  22. [2015]

    Temporal attention unit: Towards efficient spatiotemporal predictive learning

    [Tan et al., 2023] Cheng Tan, Zhangyang Gao, Lirong Wu, Yongjie Xu, Jun Xia, Siyuan Li, and Stan Z Li. Temporal attention unit: Towards efficient spatiotemporal predictive learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 18...

    2023

  23. [2016]

    Get rid of isolation: A continuous multi-task spatio- temporal learning framework

    [Yi et al., 2024] Zhongchao Yi, Zhengyang Zhou, Qihe Huang, Yanjiang Chen, Liheng Yu, Xu Wang, and Yang Wang. Get rid of isolation: A continuous multi-task spatio- temporal learning framework. In The Thirty-eighth An- nual Conference on Neural Information Processing Sys- tems,

    2024

  24. [2017]

    Memory in memory: A predictive neural network for learning higher-order non-stationarity from spatiotempo- ral dynamics

    [Wang et al., 2019] Yunbo Wang, Jianjin Zhang, Hongyu Zhu, Mingsheng Long, Jianmin Wang, and Philip S Yu. Memory in memory: A predictive neural network for learning higher-order non-stationarity from spatiotempo- ral dynamics. In Proceedings of the IEEE/CVF conference on compu...

    2019

  25. [2018]

    Self-attention con- vlstm for spatiotemporal prediction

    [Lin et al., 2020] Zhihui Lin, Maomao Li, Zhuobin Zheng, Yangyang Cheng, and Chun Yuan. Self-attention con- vlstm for spatiotemporal prediction. In Proceedings of the AAAI conference on artificial intelligence, volume 34, pages 11531–11538,

    2020

  26. [2019]

    Diffstg: Probabilistic spatio-temporal graph forecasting with denoising diffusion models

    [Wen et al., 2023] Haomin Wen, Youfang Lin, Yutong Xia, Huaiyu Wan, Qingsong Wen, Roger Zimmermann, and Yuxuan Liang. Diffstg: Probabilistic spatio-temporal graph forecasting with denoising diffusion models. InPro- ceedings of the 31st ACM International Conference on Ad- vance...

    2023

  27. [2020]

    Stable video diffusion: Scaling latent video diffusion models to large datasets

    [Blattmann et al., 2023] Andreas Blattmann, Tim Dockhorn, Sumith Kulal, Daniel Mendelevitch, Maciej Kilian, Do- minik Lorenz, Yam Levi, Zion English, Vikram V oleti, Adam Letts, et al. Stable video diffusion: Scaling latent video diffusion models to large datasets. arXiv prepr...

    2023 arXiv

  28. [2021]

    How i warped your noise: a temporally-correlated noise prior for diffusion models

    [Chang et al., 2024] Pascal Chang, Jingwei Tang, Markus Gross, and Vinicius C Azevedo. How i warped your noise: a temporally-correlated noise prior for diffusion models. In The Twelfth International Conference on Learning Rep- resentations,

    2024

  29. [2022]

    Time-llm: Time series forecasting by reprogramming large language mod- els

    [Jin et al., ] Ming Jin, Shiyu Wang, Lintao Ma, Zhixuan Chu, James Y Zhang, Xiaoming Shi, Pin-Yu Chen, Yuxuan Liang, Yuan-Fang Li, Shirui Pan, et al. Time-llm: Time series forecasting by reprogramming large language mod- els. In The Twelfth International Conference on Learning...

    2018

  30. [2023]

    Mau: A motion-aware unit for video prediction and be- yond

    [Chang et al., 2021] Zheng Chang, Xinfeng Zhang, Shanshe Wang, Siwei Ma, Yan Ye, Xiang Xinguang, and Wen Gao. Mau: A motion-aware unit for video prediction and be- yond. Advances in Neural Information Processing Sys- tems, 34:26950–26962,

    2021

  31. [2024]

    St-norm: Spatial and tem- poral normalization for multi-variate time series forecast- ing

    [Deng et al., 2021] Jinliang Deng, Xiusi Chen, Renhe Jiang, Xuan Song, and Ivor W Tsang. St-norm: Spatial and tem- poral normalization for multi-variate time series forecast- ing. In Proceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining , pages 269–278,

    2021

Pith tools

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