pith. machine review for the scientific record. sign in

arxiv: 2604.08199 · v1 · submitted 2026-04-09 · 💻 cs.NI

Recognition: 2 theorem links

· Lean Theorem

Beyond Static Forecasting: Unleashing the Power of World Models for Mobile Traffic Extrapolation

Xiaoqian Qi , Haoye Chai , Yue Wang , Yong Li
Authors on Pith no claims yet

Pith reviewed 2026-05-10 17:41 UTC · model grok-4.3

classification 💻 cs.NI
keywords mobile traffic predictionworld modelsnetwork parameter adjustmentcounterfactual simulationmultimodal fusiondigital twinreinforcement learning
0
0 comments X

The pith

MobiWM learns mobile traffic dynamics under continuous network parameter changes to support unlimited-horizon counterfactual simulations.

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

The paper seeks to replace static long-term traffic forecasts with a world model that treats traffic volume as the system state and explicitly learns its response to actions such as power, azimuth, and tilt adjustments. By encoding those actions together with fused image and sequence context, the model produces future state predictions that can be rolled out for arbitrary numbers of steps. This construction supplies operators an interactive simulation space in which different adjustment sequences can be tested without altering live networks. A reader would care because it converts passive prediction into an active planning tool that supports optimization loops. Experiments on data from over thirty thousand cells show the resulting distributions match observed traffic more closely than prior methods across multiple scenarios.

Core claim

Taking mobile traffic as the system state, MobiWM models the dynamics between the states and network parameter actions, including power, azimuth, mechanical tilt, and electrical tilt through a predictive backbone. It fuses multimodal environmental contexts, comprising both image and sequential data, with encoded actions, leveraging shared spatial semantics to enhance spatial understanding. Leveraging the capacity of world models to capture real-world operational dynamics, MobiWM supports unlimited-horizon rollout over continuous network-adjustment action trajectories, providing operators with an explorable counterfactual simulation environment for network planning and optimization.

What carries the argument

The predictive backbone that models traffic state transitions under encoded network actions, augmented by multimodal fusion of image and sequence data that shares spatial semantics.

If this is right

  • Operators obtain an interactive environment for testing arbitrary sequences of network adjustments before deployment.
  • Reinforcement learning agents can be trained directly inside the model to discover better parameter policies.
  • Digital-twin management of wireless networks becomes feasible through repeated high-fidelity rollouts.
  • The same architecture can be applied to other systems whose states change under controllable parameters.

Where Pith is reading between the lines

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

  • The simulation capability could shorten the cycle of network planning by replacing some field trials with model-based exploration.
  • Connecting the model to live telemetry streams would allow continuous recalibration as conditions evolve.
  • The approach suggests a template for building world models in other infrastructure domains that combine spatial imagery with time-series measurements.

Load-bearing premise

Dynamics learned from historical variable-parameter data will generalize accurately to unseen real-world operational conditions, and multimodal fusion of images and sequences captures all relevant spatial factors without omission.

What would settle it

Train MobiWM on traffic and adjustment records from eight districts, then evaluate its rollout accuracy on the ninth district using previously unseen sequences of power and tilt changes; the claim fails if the generated traffic distributions diverge sharply from measured values.

Figures

Figures reproduced from arXiv: 2604.08199 by Haoye Chai, Xiaoqian Qi, Yong Li, Yue Wang.

Figure 1
Figure 1. Figure 1: Comparison of the traditional static mobile traffic prediction models and the proposed mobile network world model [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Diagram of the graph batch and cell mask for irreg [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the mobile network world model, Mo [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: View of the variable-parameter mobile traffic [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Ablation study on environment context modalities [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Performance comparison during emergency events. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Model efficiency comparison. Bubble position en [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
read the original abstract

Mobile traffic prediction is a fundamental yet challenging problem for wireless network planning and optimization. Existing models focus on learning static long-term temporal patterns in mobile traffic series, which limits their ability to capture the dynamics between mobile traffic and network parameter adjustments. In this paper, we propose MobiWM, a world model for mobile networks. Taking mobile traffic as the system state, MobiWM models the dynamics between the states and network parameter actions, including power, azimuth, mechanical tilt, and electrical tilt through a predictive backbone. It fuses multimodal environmental contexts, comprising both image and sequential data, with encoded actions, leveraging shared spatial semantics to enhance spatial understanding. Leveraging the capacity of world models to capture real-world operational dynamics, MobiWM supports unlimited-horizon rollout over continuous network-adjustment action trajectories, providing operators with an explorable counterfactual simulation environment for network planning and optimization. Extensive experiments on variable-parameter mobile traffic data covering 31,900 cells across 9 districts demonstrate that MobiWM achieves the best distributional fidelity across all evaluation scenarios, significantly outperforming existing traffic prediction baselines and representative world models. A downstream RL-based case study further validates MobiWM as a simulation environment for network optimization, establishing a new paradigm for digital twin-driven wireless network management.

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 / 1 minor

Summary. The paper proposes MobiWM, a world model for mobile networks that treats mobile traffic as the system state and models its dynamics with network parameter actions (power, azimuth, mechanical tilt, electrical tilt) via a predictive backbone. Multimodal environmental contexts (images and sequences) are fused with encoded actions to enhance spatial understanding. The approach supports unlimited-horizon rollouts over continuous action trajectories for counterfactual simulation in network planning. Experiments on variable-parameter data from 31,900 cells across 9 districts claim superior distributional fidelity over traffic prediction baselines and world models, with a downstream RL case study validating its use as a simulation environment.

Significance. If substantiated, the work could advance digital-twin approaches in wireless networks by shifting from static forecasting to action-conditioned, long-horizon simulation. This would enable operators to explore optimization trajectories in a learned dynamics model rather than relying on short-term predictors or manual tuning.

major comments (3)
  1. Abstract: The central performance claim that MobiWM 'achieves the best distributional fidelity across all evaluation scenarios' and 'significantly outperforming' baselines supplies no quantitative metrics, error bars, baseline implementation details, training procedure, or comparison tables, so the claim cannot be evaluated.
  2. Evaluation section (implied by abstract): The headline assertion of unlimited-horizon rollout over arbitrary continuous action trajectories (power, azimuth, tilts) requires that the learned transition function generalizes to unseen action ranges; no explicit OOD validation, held-out action trajectories, or ablation on extrapolation is described, which is load-bearing for the counterfactual simulation use case.
  3. Abstract / method description: The multimodal fusion of images and sequences is asserted to capture all relevant spatial factors via shared semantics, yet no ablation study, sensitivity analysis, or completeness check is referenced to confirm that no critical spatial factors are omitted.
minor comments (1)
  1. The abstract would be strengthened by referencing at least one key quantitative result or table to ground the distributional-fidelity claim.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the detailed and constructive comments on our manuscript. We have carefully considered each point and provide our responses below, indicating the revisions we plan to make.

read point-by-point responses

  1. Referee: Abstract: The central performance claim that MobiWM 'achieves the best distributional fidelity across all evaluation scenarios' and 'significantly outperforming' baselines supplies no quantitative metrics, error bars, baseline implementation details, training procedure, or comparison tables, so the claim cannot be evaluated.

    Authors: We agree that the abstract, being a concise summary, does not include specific quantitative details. The detailed metrics, including distributional fidelity scores, comparisons with baselines, and implementation details, are provided in the Evaluation section of the manuscript. To make the abstract more informative and allow direct evaluation of the claims, we will revise it to include key quantitative results from our experiments, such as the reported improvements in fidelity metrics. revision: yes

  2. Referee: Evaluation section (implied by abstract): The headline assertion of unlimited-horizon rollout over arbitrary continuous action trajectories (power, azimuth, tilts) requires that the learned transition function generalizes to unseen action ranges; no explicit OOD validation, held-out action trajectories, or ablation on extrapolation is described, which is load-bearing for the counterfactual simulation use case.

    Authors: The dataset used in our experiments consists of variable-parameter mobile traffic data, which inherently includes diverse action values from real-world operations across 31,900 cells. This supports the generalization claims to some extent. However, to explicitly address the concern regarding OOD generalization for continuous action trajectories, we will add a dedicated analysis in the revised manuscript, including held-out action trajectory tests and an ablation on extrapolation performance. revision: yes

  3. Referee: Abstract / method description: The multimodal fusion of images and sequences is asserted to capture all relevant spatial factors via shared semantics, yet no ablation study, sensitivity analysis, or completeness check is referenced to confirm that no critical spatial factors are omitted.

    Authors: We recognize the value of providing empirical evidence for the effectiveness of the multimodal fusion approach. In the revised version, we will incorporate an ablation study examining the impact of each modality (images and sequences) on the model's performance, as well as a sensitivity analysis to demonstrate that the fusion captures the critical spatial factors relevant to traffic dynamics. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper proposes MobiWM as a learned world model trained on historical variable-parameter traffic data from 31,900 cells, with evaluation on separate scenarios and a downstream RL case study. No equations, self-definitional reductions, fitted parameters renamed as predictions, or load-bearing self-citations appear in the provided abstract or description. The architecture (predictive backbone, multimodal fusion) and unlimited-horizon rollout capability are presented as empirical contributions validated by outperformance metrics rather than tautological constructions. The derivation is a standard ML training/evaluation pipeline that remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The abstract provides limited technical detail; the approach rests on standard machine-learning assumptions about learnable dynamics and introduces one new model entity without additional free parameters or axioms being enumerated.

free parameters (1)
  • neural network hyperparameters
    Standard architecture and training choices for the predictive backbone and multimodal fusion components are required but not specified.
axioms (1)
  • domain assumption Historical mobile traffic data collected under varying network parameters contains sufficient information to learn generalizable state-action dynamics.
    This assumption enables the unlimited-horizon rollout capability claimed for the world model.
invented entities (1)
  • MobiWM no independent evidence
    purpose: To serve as a predictive backbone that fuses actions and multimodal contexts for counterfactual network simulation.
    The model is the central new artifact proposed in the paper.

pith-pipeline@v0.9.0 · 5527 in / 1306 out tokens · 113653 ms · 2026-05-10T17:41:01.135757+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

42 extracted references · 25 canonical work pages · 5 internal anchors

  1. [1]

    2022.Study on Channel Model for Frequencies from 0.5 to 100 GHz

    3GPP. 2022.Study on Channel Model for Frequencies from 0.5 to 100 GHz. Technical Report TR 38.901 V17.0.0. 3rd Generation Partnership Project

    2022

  2. [2]

    2023.NR; Radio Resource Control (RRC); Protocol specification

    3GPP. 2023.NR; Radio Resource Control (RRC); Protocol specification. Technical Specification TS 38.331. 3rd Generation Partnership Project

    2023

  3. [3]

    2023.NR; User Equipment (UE) procedures in idle mode and in RRC inactive state

    3GPP. 2023.NR; User Equipment (UE) procedures in idle mode and in RRC inactive state. Technical Specification TS 38.304. 3rd Generation Partnership Project

    2023

  4. [4]

    Lei Bai, Lina Yao, Can Li, Xianzhi Wang, and Can Wang. 2020. Adaptive graph convolutional recurrent network for traffic forecasting. InProceedings of the 34th International Conference on Neural Information Processing Systems(Vancouver, BC, Canada)(NIPS ’20). Curran Associates Inc., Red Hook, NY, USA, Article 1494, 12 pages

    2020

  5. [5]

    Zineddine Bettouche, Khalid Ali, Andreas Fischer, and Andreas Kassler. 2025. HiSTM: Hierarchical Spatiotemporal Mamba for Cellular Traffic Forecasting. arXiv e-prints, Article arXiv:2508.09184 (Aug. 2025), arXiv:2508.09184 pages. arXiv:2508.09184 [cs.NI] doi:10.48550/arXiv.2508.09184

    work page doi:10.48550/arxiv.2508.09184 2025

  6. [6]

    Haoye Chai, Tao Jiang, and Li Yu. 2024. Diffusion Model-based Mobile Traffic Generation with Open Data for Network Planning and Optimization. InPro- ceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining(Barcelona, Spain)(KDD ’24). Association for Computing Machinery, New York, NY, USA, 4828–4838. doi:10.1145/3637528.3671544

    work page doi:10.1145/3637528.3671544 2024

  7. [7]

    Haoye Chai, Xiaoqian Qi, and Yong Li. 2025. Spatio-Temporal Knowledge Driven Diffusion Model for Mobile Traffic Generation.IEEE Transactions on Mobile Computing24, 6 (June 2025), 4939–4956. doi:10.1109/TMC.2025.3527966

    work page doi:10.1109/tmc.2025.3527966 2025

  8. [8]

    Haoye Chai, Xiaoqian Qi, Yibo Ma, Zhaocheng Wang, Lei Yue, and Yong Li. 2026. MobiFM: A Foundation Model for Mobile Data Forecasting.IEEE Journal on Selected Areas in Communications44 (2026), 2494–2509. doi:10.1109/JSAC.2025. 3642851

    work page doi:10.1109/jsac.2025 2026

  9. [9]

    Chang Chen, Yi-Fu Wu, Jaesik Yoon, and Sungjin Ahn. 2022. TransDreamer: Reinforcement Learning with Transformer World Models.arXiv preprint arXiv:2202.09481(2022)

    work page arXiv 2022

  10. [10]

    Qingbo Du, Faming Yin, and Zongchen Li. 2020. Base station traffic prediction using XGBoost-LSTM with feature enhancement.IET Networks9, 1 (2020), 29–37. arXiv:https://ietresearch.onlinelibrary.wiley.com/doi/pdf/10.1049/iet- net.2019.0103 doi:10.1049/iet-net.2019.0103

    work page doi:10.1049/iet- 2020

  11. [11]

    David Ha and Jürgen Schmidhuber. 2018. Recurrent world models facilitate policy evolution. InProceedings of the 32nd International Conference on Neural Information Processing Systems(Montréal, Canada)(NIPS’18). Curran Associates Inc., Red Hook, NY, USA, 2455–2467

    2018

  12. [12]

    Danijar Hafner, Timothy Lillicrap, Jimmy Ba, and Mohammad Norouzi

  13. [13]

    Dream to Control: Learning Behaviors by Latent Imagination

    Dream to Control: Learning Behaviors by Latent Imagination. arXiv e-prints, Article arXiv:1912.01603 (Dec. 2019), arXiv:1912.01603 pages. arXiv:1912.01603 [cs.LG] doi:10.48550/arXiv.1912.01603

    work page internal anchor Pith review doi:10.48550/arxiv.1912.01603 1912

  14. [15]

    Danijar Hafner, Jurgis Pasukonis, Jimmy Ba, and Timothy Lillicrap. 2023. Mastering Diverse Domains through World Models.arXiv e-prints, Article arXiv:2301.04104 (Jan. 2023), arXiv:2301.04104 pages. arXiv:2301.04104 [cs.AI] doi:10.48550/arXiv.2301.04104

    work page internal anchor Pith review doi:10.48550/arxiv.2301.04104 2023

  15. [16]

    Hamilton, Rex Ying, and Jure Leskovec

    William L. Hamilton, Rex Ying, and Jure Leskovec. 2017. Inductive representation learning on large graphs. InProceedings of the 31st International Conference on Neural Information Processing Systems(Long Beach, California, USA)(NIPS’17). Curran Associates Inc., Red Hook, NY, USA, 1025–1035

    2017

  16. [17]

    Nicklas Hansen, Hao Su, and Xiaolong Wang. 2023. TD-MPC2: Scalable, Robust World Models for Continuous Control.arXiv e-prints, Article arXiv:2310.16828 (Oct. 2023), arXiv:2310.16828 pages. arXiv:2310.16828 [cs.LG] doi:10.48550/arXiv. 2310.16828

    work page internal anchor Pith review doi:10.48550/arxiv 2023

  17. [18]

    Jakob Hoydis, Faycal Ait Aoudia, Sebastian Cammerer, Merlin Nimier-David, Nikolaus Binder, Guillermo Marcus, and Alexander Keller. 2023. Sionna RT: Dif- ferentiable Ray Tracing for Radio Propagation Modeling. In2023 IEEE Globecom Workshops (GC Wkshps). 317–321. doi:10.1109/GCWkshps58843.2023.10465179

    work page doi:10.1109/gcwkshps58843.2023.10465179 2023

  18. [19]

    Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure Leskovec. 2020. Open Graph Benchmark: Datasets for Machine Learning on Graphs. InAdvances in Neural Information Processing Systems, H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin (Eds.), Vol. 33. Curran Associates, Inc., 22118–22133. http...

    2020

  19. [20]

    Yaguang Li, Rose Yu, Cyrus Shahabi, and Yan Liu. 2017. Diffusion Convolutional Recurrent Neural Network: Data-Driven Traffic Forecasting.arXiv e-prints, Arti- cle arXiv:1707.01926 (July 2017), arXiv:1707.01926 pages. arXiv:1707.01926 [cs.LG] doi:10.48550/arXiv.1707.01926

    work page doi:10.48550/arxiv.1707.01926 2017

  20. [21]

    Yong Liu, Tengge Hu, Haoran Zhang, Haixu Wu, Shiyu Wang, Lintao Ma, and Mingsheng Long. 2023. iTransformer: Inverted Transformers Are Effective for Time Series Forecasting.arXiv e-prints, Article arXiv:2310.06625 (Oct. 2023), arXiv:2310.06625 pages. arXiv:2310.06625 [cs.LG] doi:10.48550/arXiv.2310.06625

    work page internal anchor Pith review doi:10.48550/arxiv.2310.06625 2023

  21. [22]

    Jiaming Ma, Binwu Wang, Pengkun Wang, Zhengyang Zhou, Yudong Zhang, Xu Wang, and Yang Wang. 2025. MobiMixer: A Multi-Scale Spatiotemporal Mixing Model for Mobile Traffic Prediction.IEEE Transactions on Mobile Computing24, 11 (2025), 11972–11986. doi:10.1109/TMC.2025.3585007

    work page doi:10.1109/tmc.2025.3585007 2025

  22. [23]

    Ajila, Chung-Horng Lung, and Wayne Ding

    Ali Yadavar Nikravesh, Samuel A. Ajila, Chung-Horng Lung, and Wayne Ding

  23. [24]

    In2016 IEEE International Congress on Big Data (BigData Congress)

    Mobile Network Traffic Prediction Using MLP, MLPWD, and SVM. In2016 IEEE International Congress on Big Data (BigData Congress). 402–409. doi:10.1109/ BigDataCongress.2016.63

    2016

  24. [25]

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov

  25. [26]

    Proximal Policy Optimization Algorithms

    Proximal Policy Optimization Algorithms.arXiv e-prints, Article arXiv:1707.06347 (July 2017), arXiv:1707.06347 pages. arXiv:1707.06347 [cs.LG] doi:10.48550/arXiv.1707.06347

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1707.06347 2017

  26. [27]

    Xingjian SHI, Zhourong Chen, Hao Wang, Dit-Yan Yeung, Wai-kin Wong, and Wang-chun WOO. 2015. Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting. InAdvances in Neural Information Pro- cessing Systems, C. Cortes, N. Lawrence, D. Lee, M. Sugiyama, and R. Garnett (Eds.), Vol. 28. Curran Associates, Inc. https://proceedings.ne...

    2015

  27. [28]

    Xiaoming Shi, Shiyu Wang, Yuqi Nie, Dianqi Li, Zhou Ye, Qingsong Wen, and Ming Jin. 2024. Time-MoE: Billion-Scale Time Series Foundation Mod- els with Mixture of Experts.arXiv e-prints, Article arXiv:2409.16040 (Sept. 2024), arXiv:2409.16040 pages. arXiv:2409.16040 [cs.LG] doi:10.48550/arXiv.2409.16040

    work page doi:10.48550/arxiv.2409.16040 2024

  28. [29]

    Yantai Shu, Minfang Yu, Jiakun Liu, and O.W.W. Yang. 2003. Wireless traffic modeling and prediction using seasonal ARIMA models. InIEEE International Conference on Communications, 2003. ICC ’03., Vol. 3. 1675–1679 vol.3. doi:10. 1109/ICC.2003.1203886

    work page arXiv 2003

  29. [30]

    Yusuke Tashiro, Jiaming Song, Yang Song, and Stefano Ermon. 2021. CSDI: conditional score-based diffusion models for probabilistic time series imputation. InProceedings of the 35th International Conference on Neural Information Processing Systems (NIPS ’21). Curran Associates Inc., Red Hook, NY, USA, Article 1900, 13 pages

    2021

  30. [31]

    Hasita Veluri and Dilip Vasudevan. 2025. InFormer: A High-throughput, Ultra- efficient In-memory Compute-based Floating-point Arithmetic Accelerator for Transformers. InProceedings of the Great Lakes Symposium on VLSI 2025 (GLSVLSI ’25). Association for Computing Machinery, New York, NY, USA, 718–725. doi:10. 1145/3716368.3735246

    work page arXiv 2025

  31. [32]

    Xing Wang, Zhendong Wang, Kexin Yang, Zhiyan Song, Chong Bian, Junlan Feng, and Chao Deng. 2024. A Survey on Deep Learning for Cellular Traffic Prediction.Intelligent Computing3 (01 2024). doi:10.34133/icomputing.0054

    work page doi:10.34133/icomputing.0054 2024

  32. [33]

    WorldPop. 2018. WorldPop Open Population Data. https://www.worldpop.org/. School of Geography and Environmental Science, University of Southampton

    2018

  33. [34]

    Zonghan Wu, Shirui Pan, Guodong Long, Jing Jiang, and Chengqi Zhang. 2019. Graph WaveNet for Deep Spatial-Temporal Graph Modeling. InProceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, IJCAI-19. International Joint Conferences on Artificial Intelligence Organization, 1907–

    2019

  34. [35]

    doi:10.24963/ijcai.2019/264 Conference’17, July 2017, Washington, DC, USA Xiaoqian Qi, Haoye Chai, Yue Wang, and Yong Li

    work page doi:10.24963/ijcai.2019/264 2019

  35. [36]

    Marina, and Yue Wang

    Kai Xu, Rajkarn Singh, Hakan Bilen, Marco Fiore, Mahesh K. Marina, and Yue Wang. 2022. CartaGenie: Context-Driven Synthesis of City-Scale Mobile Network Traffic Snapshots. In2022 IEEE International Conference on Pervasive Computing and Communications (PerCom). 119–129. doi:10.1109/PerCom53586.2022.9762395

    work page doi:10.1109/percom53586.2022.9762395 2022

  36. [37]

    Linghua Yang, Wantong Chen, Xiaoxi He, Shuyue Wei, Yi Xu, Zimu Zhou, and Yongxin Tong. 2024. FedGTP: Exploiting Inter-Client Spatial Dependency in Federated Graph-based Traffic Prediction. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining(Barcelona, Spain)(KDD ’24). Association for Computing Machinery, New York, NY, US...

    work page doi:10.1145/3637528.3671613 2024

  37. [38]

    Bing Yu, Haoteng Yin, and Zhanxing Zhu. 2018. Spatio-Temporal Graph Con- volutional Networks: A Deep Learning Framework for Traffic Forecasting. In Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence, IJCAI-18. International Joint Conferences on Artificial Intelligence Organization, 3634–3640. doi:10.24963/ijcai.2018/505

    work page doi:10.24963/ijcai.2018/505 2018

  38. [39]

    Chaoyun Zhang, Paul Patras, and Hamed Haddadi. 2019. Deep Learning in Mobile and Wireless Networking: A Survey.IEEE Communications Surveys & Tutorials 21, 3 (2019), 2224–2287. doi:10.1109/COMST.2019.2904897

    work page doi:10.1109/comst.2019.2904897 2019

  39. [40]

    Junbo Zhang, Yu Zheng, and Dekang Qi. 2017. Deep spatio-temporal residual net- works for citywide crowd flows prediction. InProceedings of the Thirty-First AAAI Conference on Artificial Intelligence(San Francisco, California, USA)(AAAI’17). AAAI Press, 1655–1661

    2017

  40. [41]

    Shiyuan Zhang, Yilai Liu, Yuwei Du, Ruoxuan Yang, Dong In Kim, and Hongyang Du. 2026. U-MASK: User-adaptive Spatio-Temporal Masking for Personalized Mobile AI Applications.arXiv e-prints, Article arXiv:2601.06867 (Jan. 2026), arXiv:2601.06867 pages. arXiv:2601.06867 [stat.ML] doi:10.48550/arXiv.2601. 06867

    work page doi:10.48550/arxiv.2601 2026

  41. [42]

    Weipu Zhang, Gang Wang, Jian Sun, Yetian Yuan, and Gao Huang. 2023. STORM: efficient stochastic transformer based world models for reinforcement learning. In Proceedings of the 37th International Conference on Neural Information Processing Systems(New Orleans, LA, USA)(NIPS ’23). Curran Associates Inc., Red Hook, NY, USA, Article 1182, 20 pages

    2023

  42. [43]

    Changyuan Zhao, Guangyuan Liu, Ruichen Zhang, Yinqiu Liu, Jiacheng Wang, Jiawen Kang, Dusit Niyato, Zan Li, Xuemin Shen, Zhu Han, Sumei Sun, Chau Yuen, and Dong In Kim. 2026. Edge General Intelligence Through World Models, Large Language Models, and Agentic AI: Fundamentals, Solutions, and Challenges.IEEE Transactions on Cognitive Communications and Netwo...

    work page doi:10.1109/tccn.2026.3658762 2026