pith. machine review for the scientific record. sign in

arxiv: 2604.06882 · v1 · submitted 2026-04-08 · 💻 cs.RO · cs.SY· eess.SP· eess.SY

Recognition: no theorem link

Telecom World Models: Unifying Digital Twins, Foundation Models, and Predictive Planning for 6G

Anis Bara, Bohao Wang, Brahim Mefgouda, Chongwen Huang, Hang Zou, Hao Liu, Lina Bariah, Mehdi Bennis, M\'erouane Debbah, Nassim Sehad, Salma Cheour, Samson Lasaulce, Sergy Petrov, Sumudu Samarakoon, Yiwei Tao, Yuhuan Lu, Yu Tian, Yuzhi Yang

Authors on Pith no claims yet

Pith reviewed 2026-05-10 18:23 UTC · model grok-4.3

classification 💻 cs.RO cs.SYeess.SPeess.SY
keywords telecom world modelsdigital twins6G networksnetwork slicingfoundation modelspredictive planninguncertainty-aware modelingaction-conditioned simulation
0
0 comments X

The pith

A Telecom World Model decomposes networks into controllable and external worlds to enable action-conditioned KPI predictions and planning for 6G.

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

The paper argues that existing telecom management tools split into two incomplete camps: flexible but dynamics-blind language models and high-fidelity but rigid digital twins. To close the gap for 6G, where control actions produce uncertain, multi-layer effects, the authors propose the Telecom World Model. It splits the system into an operator-controlled part and an external part that includes propagation, mobility, traffic, and failures, then stacks three learned layers to forecast spatial fields, action-driven performance trajectories, and high-level intents. A network-slicing test shows the combined pipeline forecasts key metrics more accurately than single-layer baselines. If the approach holds, operators could run rapid what-if simulations and issue guarded plans that account for uncertainty instead of relying on either pure simulation or pure language generation.

Core claim

The central claim is that telecom system dynamics can be modeled through the Telecom World Model architecture, which decomposes the problem into a controllable system world of operator-configurable settings and an external world of propagation, mobility, traffic, and failures. A three-layer structure then supplies field-level spatial prediction, action-conditioned KPI trajectory forecasting, and foundation-model intent translation, jointly delivering state grounding, fast roll-outs, calibrated uncertainty, multi-timescale dynamics, model-based planning, and LLM-integrated guardrails. The network-slicing proof-of-concept demonstrates that the full pipeline outperforms single-world baselines.

What carries the argument

The three-layer Telecom World Model architecture, which separates controllable operator settings from external dynamics to learn and roll out action effects on system performance.

If this is right

  • The model supports rapid simulation of how operator actions change key performance indicators without running full physics-based simulators.
  • Calibrated uncertainty estimates allow planners to choose actions that remain reliable under variability in traffic or channel conditions.
  • Multi-timescale modeling captures both immediate effects and longer-term evolution of network states in one framework.
  • Foundation-model guardrails translate high-level intents into safe, executable control sequences while the dynamics layers handle the physics.
  • In network slicing the combined layers produce KPI trajectory forecasts that match reality more closely than isolated digital-twin or language-model baselines.

Where Pith is reading between the lines

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

  • The same decomposition could be applied to other 6G control loops such as dynamic beam management or edge-resource allocation to test whether the accuracy gains generalize.
  • Online updates from live telemetry streams might allow the world model to adapt continuously rather than relying on offline training alone.
  • Hybrid versions that inject known physics equations into the learned layers could reduce data requirements while preserving uncertainty calibration.

Load-bearing premise

Telecom dynamics can be decomposed into controllable and external components and learned through the three-layer architecture to produce accurate, calibrated predictions usable for real 6G planning.

What would settle it

In a live or emulated network-slicing deployment, compare the full TWM pipeline's predicted KPI trajectories against ground-truth measurements across varied action sequences and timescales; if the predictions are no more accurate than single-world baselines or if uncertainty intervals fail to contain observed values at the claimed rate, the claim is falsified.

Figures

Figures reproduced from arXiv: 2604.06882 by Anis Bara, Bohao Wang, Brahim Mefgouda, Chongwen Huang, Hang Zou, Hao Liu, Lina Bariah, Mehdi Bennis, M\'erouane Debbah, Nassim Sehad, Salma Cheour, Samson Lasaulce, Sergy Petrov, Sumudu Samarakoon, Yiwei Tao, Yuhuan Lu, Yu Tian, Yuzhi Yang.

Figure 1
Figure 1. Figure 1: Positioning of TWM in the landscape of related [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Two-world, three-layer Telecom World Model [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: TWM layers across planning timescales. Each horizon [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Layer-specific data, alignment, training, and evaluation [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: PoC environment. (a) Shaded areas are schematic operational zones; colored cell outlines indicate region-associated [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: SLA Compliance Margin by method and scenario [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
read the original abstract

The integration of machine learning tools into telecom networks, has led to two prevailing paradigms, namely, language-based systems, such as Large Language Models (LLMs), and physics-based systems, such as Digital Twins (DTs). While LLM-based approaches enable flexible interaction and automation, they lack explicit representations of network dynamics. DTs, in contrast, offer a high-fidelity network simulation, but remain scenario-specific and are not designed for learning or decision-making under uncertainty. This gap becomes critical for 6G systems, where decisions must take into account the evolving network states, uncertainty, and the cascading effects of control actions across multiple layers. In this article, we introduce the {Telecom World Model}~(TWM) concept, an architecture for learned, action-conditioned, uncertainty-aware modeling of telecom system dynamics. We decompose the problem into two interacting worlds, a controllable system world consisting of operator-configurable settings and an external world that captures propagation, mobility, traffic, and failures. We propose a three-layer architecture, comprising a field world model for spatial environment prediction, a control/dynamics world model for action-conditioned Key Performance Indicator (KPI) trajectory prediction, and a telecom foundation model layer for intent translation and orchestration. We showcase a comparative analysis between existing paradigms, which demonstrates that TWM jointly provides telecom state grounding, fast action-conditioned roll-outs, calibrated uncertainty, multi-timescale dynamics, model-based planning, and LLM-integrated guardrails. Furthermore, we present a proof-of-concept on network slicing to validate the proposed architecture, showing that the full three-layer pipeline outperforms single-world baselines and accurately predicts KPI trajectories.

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

2 major / 1 minor

Summary. The manuscript introduces the Telecom World Model (TWM) concept as an architecture for learned, action-conditioned, uncertainty-aware modeling of telecom system dynamics in 6G networks. It decomposes the problem into a controllable system world (operator-configurable settings) and an external world (propagation, mobility, traffic, failures), proposing a three-layer architecture: a field world model for spatial environment prediction, a control/dynamics world model for action-conditioned KPI trajectory prediction, and a telecom foundation model for intent translation and orchestration. The paper presents a comparative analysis of existing paradigms (LLMs and digital twins) and a proof-of-concept on network slicing claiming that the full three-layer pipeline outperforms single-world baselines and accurately predicts KPI trajectories.

Significance. If the architecture and PoC results hold under rigorous validation, TWM could meaningfully bridge the gap between physics-based digital twins and language-based foundation models by enabling fast roll-outs, calibrated uncertainty, multi-timescale dynamics, model-based planning, and LLM guardrails. This would represent a substantive contribution to predictive network management for 6G, particularly if the decomposition into controllable/external worlds proves separable and learnable at scale.

major comments (2)
  1. [proof-of-concept on network slicing] Proof-of-concept on network slicing: the claim that the full three-layer pipeline outperforms single-world baselines and accurately predicts KPI trajectories is asserted without reporting of experimental details, including datasets, specific quantitative metrics (e.g., prediction error, proper scoring rules for uncertainty calibration), baselines, error bars, or confirmation that all three layers (including the foundation model for intent translation) were active. This gap prevents assessment of whether the results support the full set of asserted properties such as calibrated uncertainty and multi-timescale dynamics.
  2. [three-layer architecture description] The separability assumption underlying the controllable/external world decomposition is central to the architecture but remains high-level; no explicit learning procedures, training objectives, or validation that the decomposition yields accurate action-conditioned predictions are provided to substantiate that the three-layer design delivers the claimed advantages over existing paradigms.
minor comments (1)
  1. [abstract] The abstract contains LaTeX-style braces around 'Telecom World Model' that appear to be a formatting artifact and should be cleaned for readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the constructive and insightful review. We address each major comment point-by-point below and will revise the manuscript to incorporate the requested details, thereby strengthening the presentation of the proof-of-concept and architecture.

read point-by-point responses

  1. Referee: Proof-of-concept on network slicing: the claim that the full three-layer pipeline outperforms single-world baselines and accurately predicts KPI trajectories is asserted without reporting of experimental details, including datasets, specific quantitative metrics (e.g., prediction error, proper scoring rules for uncertainty calibration), baselines, error bars, or confirmation that all three layers (including the foundation model for intent translation) were active. This gap prevents assessment of whether the results support the full set of asserted properties such as calibrated uncertainty and multi-timescale dynamics.

    Authors: We acknowledge that the current manuscript presents the network slicing proof-of-concept at a high level to illustrate the overall architecture, without the granular experimental details needed for full reproducibility and assessment. In the revised version, we will expand the relevant section to report: the specific dataset (synthetic traces generated from a 6G network simulator incorporating realistic mobility, traffic, and propagation models); quantitative metrics including RMSE and MAE for KPI trajectory prediction error, CRPS and calibration error for uncertainty quantification; the full set of baselines (single-layer field model, dynamics model alone, LLM-only predictor, and physics-based DT); error bars and statistical significance from multiple independent runs; and explicit confirmation that the complete three-layer pipeline was active, with the foundation model performing intent-to-action translation. These additions will directly support evaluation of calibrated uncertainty and multi-timescale dynamics. revision: yes

  2. Referee: The separability assumption underlying the controllable/external world decomposition is central to the architecture but remains high-level; no explicit learning procedures, training objectives, or validation that the decomposition yields accurate action-conditioned predictions are provided to substantiate that the three-layer design delivers the claimed advantages over existing paradigms.

    Authors: The controllable/external world decomposition is motivated by the fundamental distinction in telecom systems between operator actions and exogenous dynamics, but we agree the manuscript leaves the learning and validation aspects at a conceptual level. In revision, we will add a dedicated subsection on implementation details, including: training objectives for the field world model (supervised spatial prediction loss on environmental data), the control/dynamics model (action-conditioned sequence prediction with combined regression and uncertainty losses, e.g., via probabilistic outputs), and the foundation model (fine-tuning for intent parsing and orchestration); explicit learning procedures such as end-to-end or staged training; and validation methods (e.g., held-out trajectory testing and ablation studies confirming improved action-conditioned accuracy and separability benefits over monolithic baselines). This will more rigorously substantiate the claimed advantages. revision: yes

Circularity Check

0 steps flagged

Conceptual architecture proposal with empirical PoC shows no derivation circularity

full rationale

The paper is a forward-looking conceptual framework introducing the TWM architecture, its decomposition into controllable/external worlds, and a three-layer structure (field world model, control/dynamics world model, telecom foundation model). The central claims rest on this proposed unification and a network-slicing proof-of-concept that reports outperformance over baselines plus KPI trajectory prediction. No equations, fitted parameters, or mathematical derivations appear that reduce predictions to inputs by construction. No self-citations serve as load-bearing uniqueness theorems, no ansatzes are smuggled, and no known results are merely renamed. The work is self-contained as an architectural proposal whose validation is empirical rather than tautological.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the domain assumption that telecom dynamics can be decomposed into controllable and external worlds and modeled in a learned, action-conditioned manner; no free parameters or invented entities with independent evidence are specified in the abstract.

axioms (1)
  • domain assumption Telecom networks can be decomposed into a controllable system world of operator settings and an external world of propagation, mobility, traffic, and failures.
    This decomposition is explicitly proposed as the basis for the TWM architecture in the abstract.
invented entities (1)
  • Telecom World Model (TWM) no independent evidence
    purpose: Learned, action-conditioned, uncertainty-aware modeling of telecom system dynamics via a three-layer architecture.
    Newly introduced concept; the abstract provides no independent evidence outside the high-level POC description.

pith-pipeline@v0.9.0 · 5685 in / 1537 out tokens · 90997 ms · 2026-05-10T18:23:24.704308+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

65 extracted references · 18 canonical work pages · 4 internal anchors

  1. [1]

    Language Models are Few-Shot Learners,

    T. B. Brownet al., “Language Models are Few-Shot Learners,” in Advances in Neural Information Processing Systems (NeurIPS), 2020

    2020

  2. [2]

    Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities

    G. Comaniciet al., “Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities,”arXiv preprint arXiv:2507.06261, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  3. [3]

    Large language models for networking: Applications, enabling techniques, and challenges,

    Z. Zhanget al., “Large Language Models for Networking: Appli- cations, Challenges, and the ChatNet Framework,”arXiv preprint arXiv:2311.17474, 2023

    work page arXiv 2023

  4. [4]

    Large Generative AI Models for Telecom: The Next Big Thing?

    L. Bariah, Q. Zhao, H. Zou, Y . Tian, F. Bader, and M. Debbah, “Large Generative AI Models for Telecom: The Next Big Thing?”IEEE Communications Magazine, vol. 62, no. 11, pp. 84–90, 2024

    2024

  5. [5]

    TelecomGPT: A Framework to Build Telecom-Specific Large Language Models,

    H. Zou, Q. Zhao, Y . Tian, L. Bariah, F. Bader, T. Lestable, and M. Debbah, “TelecomGPT: A Framework to Build Telecom-Specific Large Language Models,”IEEE Transactions on Machine Learning in Communications and Networking, vol. 3, pp. 948–975, 2025

    2025

  6. [6]

    Large language models in 6g from standard to on-device networks,

    H. Zou, Q. Zhao, S. Lasaulceet al., “Large language models in 6g from standard to on-device networks,”Nature Reviews Electrical Engineering, vol. 3, pp. 123–134, 2026

    2026

  7. [7]

    Mastering diverse control tasks through world models,

    D. Hafner, J. Pasukonis, J. Ba, and T. Lillicrap, “Mastering diverse control tasks through world models,”Nature, vol. 640, no. 7839, pp. 647–653, 2025

    2025

  8. [8]

    Sora: Creating video from text,

    OpenAI, “Sora: Creating video from text,” OpenAI Blog, 2024

    2024

  9. [9]

    A Path Towards Autonomous Machine Intelligence,

    Y . LeCun, “A Path Towards Autonomous Machine Intelligence,” Courant Institute of Mathematical Sciences, New York University, Tech. Rep., 2022, version 0.9.2, OpenReview preprint

    2022

  10. [10]

    Single and Multi-Agent Deep Reinforce- ment Learning for AI-Enabled Wireless Networks: A Tutorial,

    A. Feriani and E. Hossain, “Single and Multi-Agent Deep Reinforce- ment Learning for AI-Enabled Wireless Networks: A Tutorial,”IEEE Communications Surveys & Tutorials, vol. 23, no. 2, pp. 1226–1252, 2021

    2021

  11. [11]

    Model-based deep reinforcement learning framework for channel access in wireless networks,

    J. I. Park, J. B. Chae, and K. W. Choi, “Model-based deep reinforcement learning framework for channel access in wireless networks,”IEEE Internet of Things Journal, vol. 11, no. 6, pp. 10 150–10 167, 2024

    2024

  12. [12]

    MobiWorld: World Models for Mobile Wireless Network,

    H. Chai, Y . Yuan, and Y . Li, “MobiWorld: World Models for Mobile Wireless Network,”arXiv preprint arXiv:2507.09462, 2025

    work page arXiv 2025

  13. [13]

    Wireless Large AI Model: Shaping the AI-Native Future of 6G and Beyond

    F. Zhuet al., “Wireless large ai model: Shaping the ai-native future of 6g and beyond,”arXiv preprint arXiv:2504.14653, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  14. [14]

    Large-scale AI in telecom: Charting the roadmap for innovation, scalability, and enhanced digital experiences,

    A. Shahidet al., “Large-scale AI in telecom: Charting the roadmap for innovation, scalability, and enhanced digital experiences,”arXiv preprint arXiv:2503.04184, 2025

    work page arXiv 2025

  15. [15]

    Recurrent World Models Facilitate Policy Evolution,

    D. Ha and J. Schmidhuber, “Recurrent World Models Facilitate Policy Evolution,” inAdvances in Neural Information Processing Systems (NeurIPS), 2018

    2018

  16. [16]

    Learning Latent Dynamics for Planning from Pixels,

    D. Hafner, T. Lillicrap, I. Fischer, R. Villegas, D. Ha, H. Lee, and J. Davidson, “Learning Latent Dynamics for Planning from Pixels,” in International Conference on Machine Learning (ICML), 2019. 19

    2019

  17. [17]

    Dream to Control: Learning Behaviors by Latent Imagination,

    D. Hafner, T. Lillicrap, J. Ba, and M. Norouzi, “Dream to Control: Learning Behaviors by Latent Imagination,” inInternational Conference on Learning Representations (ICLR), 2020

    2020

  18. [18]

    Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model,

    J. Schrittwieser, I. Antonoglou, T. Hubert, K. Simonyan, L. Sifre, S. Schmitt, A. Guez, E. Lockhart, D. Hassabis, T. Graepel, T. Lillicrap, and D. Silver, “Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model,”Nature, vol. 588, pp. 604–609, 2020

    2020

  19. [19]

    Transformers are Sample- Efficient World Models,

    V . Micheli, E. Alonso, and F. Fleuret, “Transformers are Sample- Efficient World Models,” inInternational Conference on Learning Representations (ICLR), 2023

    2023

  20. [20]

    Temporal Difference Learning for Model Predictive Control,

    N. A. Hansen, H. Su, and X. Wang, “Temporal Difference Learning for Model Predictive Control,” inProceedings of the 39th International Conference on Machine Learning (ICML), vol. 162, 2022, pp. 8387– 8406

    2022

  21. [21]

    Stochastic Video Generation with a Learned Prior,

    E. Denton and R. Fergus, “Stochastic Video Generation with a Learned Prior,” inInternational Conference on Machine Learning (ICML), 2018

    2018

  22. [22]

    Learning Interactive Real-World Simulators,

    S. Yang, Y . Du, K. Ghasemipour, J. Tompson, L. Kaelbling, D. Schuur- mans, and P. Abbeel, “Learning Interactive Real-World Simulators,” in International Conference on Learning Representations (ICLR), 2024

    2024

  23. [23]

    Genie: Generative Interactive Environments,

    J. Bruce, M. D. Dennis, A. Edwards, J. Parker-Holder, Y . Shi, E. Hughes, M. Laiet al., “Genie: Generative Interactive Environments,” inProceed- ings of the 41st International Conference on Machine Learning (ICML), vol. 235, 2024, pp. 4603–4623

    2024

  24. [24]

    Diffusion models are real-time game engines

    D. Valevski, Y . Leviathan, M. Arar, and S. Fruchter, “Diffusion Models Are Real-Time Game Engines,”arXiv preprint arXiv:2408.14837, 2024

    work page arXiv 2024

  25. [25]

    Revisiting Feature Prediction for Learning Visual Representations from Video,

    A. Bardes, Q. Garrido, J. Ponce, X. Chen, M. G. Rabbat, Y . LeCun, M. Assran, and N. Ballas, “Revisiting Feature Prediction for Learning Visual Representations from Video,”Transactions on Machine Learning Research, 2024

    2024

  26. [26]

    3D-VLA: A 3D Vision-Language-Action Generative World Model,

    H. Zhen, X. Qiu, P. Chen, J. Yang, X. Yan, Y . Du, Y . Hong, and C. Gan, “3D-VLA: A 3D Vision-Language-Action Generative World Model,” in Proceedings of the 41st International Conference on Machine Learning (ICML), vol. 235. PMLR, 2024, pp. 61 229–61 245

    2024

  27. [27]

    WorldVLA: Towards Autoregressive Action World Model

    J. Cen, C. Yu, H. Yuan, Y . Jiang, S. Huang, J. Guo, X. Li, Y . Song, H. Luo, F. Wang, D. Zhao, and H. Chen, “WorldVLA: Towards Autoregressive Action World Model,”arXiv preprint arXiv:2506.21539, 2025

    work page internal anchor Pith review arXiv 2025

  28. [28]

    OccWorld: Learning a 3D Occupancy World Model for Autonomous Driving,

    W. Zheng, W. Chen, Y . Huang, B. Zhang, Y . Duan, and J. Lu, “OccWorld: Learning a 3D Occupancy World Model for Autonomous Driving,” inEuropean Conference on Computer Vision (ECCV), 2024, pp. 55–72

    2024

  29. [29]

    Geodrive: 3d geometry-informed driving world model with precise action control.arXiv preprint arXiv:2505.22421, 2025

    A. Chen, W. Zheng, Y . Wang, X. Zhang, K. Zhan, P. Jia, K. Keutzer, and S. Zhang, “GeoDrive: 3D Geometry-Informed Driving World Model with Precise Action Control,”arXiv preprint arXiv:2505.22421, 2025

    work page arXiv 2025

  30. [30]

    Recommendation ITU-T Y .3090: Digital twin network – Re- quirements and architecture,

    ITU-T, “Recommendation ITU-T Y .3090: Digital twin network – Re- quirements and architecture,” International Telecommunication Union, 2022

    2022

  31. [31]

    Digital- Twin-Enabled 6G: Vision, Architectural Trends, and Future Directions,

    L. U. Khan, W. Saad, D. Niyato, Z. Han, and C. S. Hong, “Digital- Twin-Enabled 6G: Vision, Architectural Trends, and Future Directions,” IEEE Communications Magazine, vol. 60, no. 1, pp. 74–80, 2022

    2022

  32. [32]

    Network Digital Twin: Context, Enabling Tech- nologies, and Opportunities,

    P. Almasanet al., “Network Digital Twin: Context, Enabling Tech- nologies, and Opportunities,”IEEE Communications Magazine, vol. 60, no. 11, pp. 22–27, 2022

    2022

  33. [33]

    Digital Twin of Wireless Systems: Overview, Taxonomy, Challenges, and Opportunities,

    L. U. Khan, Z. Han, W. Saad, E. Hossain, M. Guizani, and C. S. Hong, “Digital Twin of Wireless Systems: Overview, Taxonomy, Challenges, and Opportunities,”IEEE Communications Surveys & Tutorials, vol. 24, no. 4, pp. 2230–2254, 2022

    2022

  34. [34]

    Large Language Model (LLM) for Telecommunications: A Comprehensive Survey on Principles, Key Techniques, and Opportu- nities,

    H. Zhouet al., “Large Language Model (LLM) for Telecommunications: A Comprehensive Survey on Principles, Key Techniques, and Opportu- nities,”IEEE Communications Surveys & Tutorials, vol. 27, no. 3, pp. 1955–2005, 2025

    1955

  35. [35]

    NetGPT: An AI-Native Network Architecture for Provisioning Beyond Personalized Generative Services,

    Y . Chen, R. Li, Z. Zhao, C. Peng, J. Wu, E. Hossain, and H. Zhang, “NetGPT: An AI-Native Network Architecture for Provisioning Beyond Personalized Generative Services,”IEEE Network, vol. 38, no. 6, pp. 404–413, 2024

    2024

  36. [36]

    WirelessLLM: Empowering large language models towards wireless intelligence,

    J. Shaoet al., “WirelessLLM: Empowering Large Language Models Towards Wireless Intelligence,”arXiv preprint arXiv:2405.17053, 2024

    work page arXiv 2024

  37. [37]

    RF-GPT: Teaching AI to see the wireless world,

    H. Zou, Y . Tian, B. Wang, L. Bariah, S. Lasaulce, C. Huang, and M. Debbah, “RF-GPT: Teaching AI to See the Wireless World,”arXiv preprint arXiv:2602.14833, 2026

    work page arXiv 2026

  38. [38]

    A Survey on Large Lan- guage Models for Communication, Network, and Service Management: Application Insights, Challenges, and Future Directions,

    G. O. Boateng, H. Sami, A. Alaghaet al., “A Survey on Large Lan- guage Models for Communication, Network, and Service Management: Application Insights, Challenges, and Future Directions,”arXiv preprint arXiv:2412.19823, 2024

    work page arXiv 2024

  39. [39]

    6g comprehensive intelligence: Network operations and optimization based on large language models,

    S. Long, F. Tang, Y . Li, T. Tan, Z. Jin, M. Zhao, and N. Kato, “6g comprehensive intelligence: Network operations and optimization based on large language models,”IEEE Network, vol. 39, no. 4, pp. 192–201, 2025

    2025

  40. [40]

    Large language models for wireless communications: From adaptation to autonomy,

    L. Liang, H. Ye, Y . Sheng, O. Wang, J. Wang, S. Jin, and G. Y . Li, “Large language models for wireless communications: From adaptation to autonomy,”IEEE Communications Magazine, 2026

    2026

  41. [41]

    TeleQnA: A Benchmark Dataset to Assess Large Language Models Telecommunications Knowledge,

    A. Maatouk, F. Ayed, N. Piovesan, A. De Domenico, M. Debbah, and Z.-Q. Luo, “TeleQnA: A Benchmark Dataset to Assess Large Language Models Telecommunications Knowledge,”IEEE Network, 2024

    2024

  42. [42]

    Learning Optimal Resource Allocations in Wireless Systems,

    M. Eisen, C. Zhang, L. F. O. Chamon, D. D. Lee, and A. Ribeiro, “Learning Optimal Resource Allocations in Wireless Systems,”IEEE Transactions on Signal Processing, vol. 67, no. 10, pp. 2775–2790, 2019

    2019

  43. [43]

    Learning latent multi- modal dynamics for optimized resource planning,

    C. Bou Chaaya, A. M. Girgis, and M. Bennis, “Learning latent multi- modal dynamics for optimized resource planning,”IEEE Transactions on Wireless Communications, vol. 25, pp. 9591–9607, 2026

    2026

  44. [44]

    Radiounet: Fast radio map estimation with convolutional neural networks,

    R. Levie, C ¸ . Yapar, G. Kutyniok, and G. Caire, “Radiounet: Fast radio map estimation with convolutional neural networks,”IEEE Transactions on Wireless Communications, vol. 20, no. 6, pp. 4001–4015, 2021

    2021

  45. [45]

    PMNet: Robust Pathloss Map Prediction via Supervised Learning,

    J.-H. Lee, O. G. Serbetci, D. P. Selvam, and A. F. Molisch, “PMNet: Robust Pathloss Map Prediction via Supervised Learning,” inIEEE Global Communications Conference (GLOBECOM), 2023

    2023

  46. [46]

    Deep Completion Autoencoders for Radio Map Estimation,

    Y . Teganya and D. Romero, “Deep Completion Autoencoders for Radio Map Estimation,”IEEE Transactions on Wireless Communications, vol. 21, no. 3, pp. 1710–1724, 2022

    2022

  47. [47]

    WiNeRT: Towards Neural Ray Tracing for Wireless Channel Modelling and Differentiable Simulations,

    T. Orekondy, P. Kumar, S. Kadambi, H. Ye, J. Soriaga, and A. Be- hboodi, “WiNeRT: Towards Neural Ray Tracing for Wireless Channel Modelling and Differentiable Simulations,” inInternational Conference on Learning Representations (ICLR), 2023

    2023

  48. [48]

    RouteNet- Erlang: A Graph Neural Network for Network Performance Evaluation,

    M. Ferriol-Galm ´es, K. Rusek, J. Su ´arez-Varela, S. Xiao, X. Shi, X. Cheng, B. Wu, P. Barlet-Ros, and A. Cabellos-Aparicio, “RouteNet- Erlang: A Graph Neural Network for Network Performance Evaluation,” inIEEE INFOCOM 2022-IEEE Conference on Computer Communica- tions, 2022, pp. 2018–2027

    2022

  49. [49]

    Sionna: An Open-Source Library for Next-Generation Physical Layer Research,

    J. Hoydis, S. Cammerer, F. A. Aoudia, A. Vem, N. Binder, G. Marcus, and A. Keller, “Sionna: An open-source library for next-generation physical layer research,”arXiv preprint arXiv:2203.11854, 2022

    work page arXiv 2022

  50. [50]

    Sionna rt: Differentiable ray tracing for radio propagation modeling,

    J. Hoydis, F. A ¨ıt Aoudia, S. Cammerer, M. Nimier-David, N. Binder, G. Marcus, and A. Keller, “Sionna rt: Differentiable ray tracing for radio propagation modeling,” in2023 IEEE Globecom Workshops (GC Wkshps). IEEE, 2023, pp. 317–321

    2023

  51. [51]

    Sionna RT: Technical report,

    F. A. Aoudia, J. Hoydis, M. Nimier-David, B. Nicolet, S. Cam- merer, and A. Keller, “Sionna rt: Technical report,”arXiv preprint arXiv:2504.21719, 2025

    work page arXiv 2025

  52. [52]

    Aerial omniverse digital twin,

    NVIDIA, “Aerial omniverse digital twin,” Apr. 2026, [Online]. Avail- able: https://docs.nvidia.com/aerial/aerial-dt/. [Accessed: Apr. 7, 2026]

    2026

  53. [53]

    Deepverse 6g: A framework for synthetic multi-modal sensing and communication datasets,

    U. Demirhan, A. Taha, and A. Alkhateeb, “Deepverse 6g: A framework for synthetic multi-modal sensing and communication datasets,”arXiv preprint, 2022

    2022

  54. [54]

    Safeslice: Enabling sla-compliant o-ran slicing via safe deep reinforcement learning,

    A. M. Nagib, H. Abou-Zeid, and H. S. Hassanein, “SafeSlice: Enabling SLA-compliant O-RAN slicing via safe deep reinforcement learning,” arXiv preprint arXiv:2503.12753, 2025

    work page arXiv 2025

  55. [55]

    Flexible RAN slicing in open RAN with constrained multi-agent reinforcement learning,

    M. Zangooei, M. Golkarifard, M. Rouili, N. Saha, and R. Boutaba, “Flexible RAN slicing in open RAN with constrained multi-agent reinforcement learning,”IEEE Journal on Selected Areas in Commu- nications, vol. 42, no. 2, pp. 280–294, 2024

    2024

  56. [56]

    Real: Reinforcement learning-enabled xapps for experimental closed-loop optimization in o-ran with osc ric and srsran,

    R. Barker, W. Saad, C. G. Brinton, and N. Soltani, “REAL: Re- inforcement learning-enabled xapps for experimental closed-loop op- timization in O-RAN with OSC RIC and srsRAN,”arXiv preprint arXiv:2502.00715, 2025

    work page arXiv 2025

  57. [57]

    ORANSlice: An open-source 5g network slicing platform for O-RAN,

    H. Cheng, R. Ravindran, M. Polese, J. Black, T. Melodia, and D. Villa, “ORANSlice: An open-source 5g network slicing platform for O-RAN,” inProceedings of the ACM MobiCom Workshop on Open RAN, 2024

    2024

  58. [58]

    Intelligent closed-loop RAN control with xapps in OpenRAN Gym,

    L. Bonati, M. Polese, S. D’Oro, S. Basagni, and T. Melodia, “Intelligent closed-loop RAN control with xapps in OpenRAN Gym,” inIEEE European Wireless Conference, 2023

    2023

  59. [59]

    Dynamic programming for pomdps using a factored state representation,

    E. A. Hansen and Z. Feng, “Dynamic programming for pomdps using a factored state representation,” inProceedings of the Fifth International Conference on Artificial Intelligence Planning Systems, 2000, pp. 130– 139

    2000

  60. [60]

    Reading radio from camera: Visually-grounded, lightweight, and interpretable RSSI prediction,

    S. Yan, T. Hu, B. Mefgouda, S. Lasaulce, and M. Debbah, “Reading radio from camera: Visually-grounded, lightweight, and interpretable rssi prediction,”arXiv preprint arXiv:2510.25936, 2025

    work page arXiv 2025

  61. [61]

    MIMO Channel Estimation Using Score- Based Generative Models,

    M. Arvinte and J. I. Tamir, “MIMO Channel Estimation Using Score- Based Generative Models,”IEEE Transactions on Wireless Communi- cations, vol. 22, no. 6, pp. 3698–3713, 2023

    2023

  62. [62]

    Generative diffusion models for high dimensional channel estimation,

    X. Zhou, L. Liang, J. Zhang, P. Jiang, Y . Li, and S. Jin, “Generative diffusion models for high dimensional channel estimation,”IEEE Trans- 20 actions on Wireless Communications, vol. 24, no. 7, pp. 5840–5854, 2025

    2025

  63. [63]

    Generative diffusion receivers: Achieving pilot-efficient mimo-ofdm communications,

    Y . Yang, O. Alhussein, A. H. Arani, Z. Zhang, and M. Debbah, “Generative diffusion receivers: Achieving pilot-efficient mimo-ofdm communications,”IEEE Transactions on Network Science and Engi- neering, vol. 13, pp. 6721–6736, 2026

    2026

  64. [64]

    Diffusion models for wireless transceivers: From pilot-efficient channel estimation to AI-native 6G receivers,

    Y . Yang, S. Yan, W. Zhou, B. Mefgouda, R. Li, Z. Zhang, and M. Debbah, “Diffusion models for wireless transceivers: From pilot- efficient channel estimation to AI-native 6G receivers,”arXiv preprint arXiv:2510.24495, 2025

    work page arXiv 2025

  65. [65]

    Qwen2.5 Technical Report

    A. Yanget al., “Qwen2. 5 technical report,”arXiv preprint arXiv:2412.15115, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024