Pith. sign in

REVIEW 4 major objections 5 minor 2 cited by

This survey argues that explicit reasoning—prompting, architecture, and learning—can replace black-box deep learning in wireless networks, and provides a layer-by-layer map of how.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 19:30 UTC pith:GOZSBBU2

load-bearing objection A useful, readable survey of AI reasoning for wireless that organizes recent work layer-by-layer, but the conclusion overclaims coverage and the real-time deployment gap is unresolved. the 4 major comments →

arxiv 2509.09193 v1 pith:GOZSBBU2 submitted 2025-09-11 cs.NI

AI Reasoning for Wireless Communications and Networking: A Survey and Perspectives

Haoxiang Luo , Yu Yan , Yanhui Bian , Wenjiao Feng , Ruichen Zhang , Yinqiu Liu , Jiacheng Wang , Gang Sun
show 4 more authors
Dusit Niyato Hongfang Yu Abbas Jamalipour Shiwen Mao
This is my paper
classification cs.NI
keywords AI reasoninglarge language modelswireless networkschain-of-thoughtretrieval-augmented generationnetwork optimization6Gsemantic communication
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper argues that conventional deep learning in wireless networks is a closed box: it cannot explain decisions, adapt to unseen conditions, or incorporate domain knowledge. The authors claim that adding explicit reasoning—prompting strategies, architectural additions, and learning paradigms—lets AI decompose network problems into transparent steps, and they organize the field by the five protocol layers. The survey's distinctive contribution is this layered map: physical, data link, network, transport, and application layers each get their challenges matched with a reasoning solution and typical scenarios. If the central claim holds, the path to 6G shifts from bigger black-box models to smaller, explainable, reasoning-enabled agents that can manage networks with minimal human oversight.

Core claim

On its own terms, the paper establishes a framework: 'reasoning method-function-wireless scenario' and 'challenge-AI reasoning solution-typical scenario.' It surveys existing results—CoT for beamforming power allocation, GNNs for link scheduling and scaling to thousands of users, RAG for modulation recognition and error correction, neuro-symbolic methods for semantic communication, and reinforcement-learning-based causal reasoning for QoE—and reads them as evidence that reasoning-enabled AI can outperform both traditional algorithms and black-box learning. The paper does not report new experiments; it synthesizes proof-of-concept results into a research agenda for AI reasoning across all lay

What carries the argument

The carrying structure is a two-part taxonomy. The first part classifies reasoning methods into three categories: prompting strategies (CoT, ToT, self-consistency), architectural approaches (ReAct, PAL, RAG, GNN, neuro-symbolic, multi-LLM, memory), and learning paradigms (self-refinement, CoT distillation, RLHF). The second part classifies tasks by function: generation, classification, optimization, prediction. The survey then maps these onto a layer-by-layer examination of the wireless protocol stack, so each layer's challenges are paired with a reasoning solution and a typical scenario. This mapping is what makes the survey actionable: it tells a researcher which reasoning tool fits which

Load-bearing premise

The load-bearing premise is that LLM-style reasoning can be made fast, cheap, and reliable enough to run within the strict real-time limits of wireless network layers—something the survey notes is not yet true for edge deployment.

What would settle it

If a straightforward benchmark showed that a CoT- or agent-based controller cannot meet the millisecond latency budget of physical-layer tasks on realistic edge hardware, or that retrieval-augmented error correction increases retransmissions rather than reducing them, the paper's central claim would be weakened. For example, run a standard beamforming or link-scheduling task with an LLM on a representative base-station processor and measure end-to-end decision time versus the deadline.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If reasoning AI delivers step-by-step decisions, network operators can verify and trust AI actions rather than accept black-box outputs.
  • CoT and PAL can replace hand-written optimization loops for physical-layer tasks like beamforming power allocation, with proof-of-concept spectral efficiency gains.
  • GNNs can make link scheduling and beamforming decentralized and scalable, e.g., a model trained on 50 users performing near-optimally on 1000 users.
  • RAG can ground LLM decisions in device manuals and standards, reducing retransmissions and enabling automatic configuration.
  • The four research directions—agentic reasoning, data-efficient learning, efficient deployment, and safety/reliability—define the agenda that must be solved before deployment.

Where Pith is reading between the lines

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

  • An implication the authors leave implicit is that the same taxonomy could double as a benchmark scheme: each layer already has concrete tasks and success metrics, enabling apples-to-apples comparison of prompting versus architectural methods.
  • The layer-by-layer treatment points toward a cross-layer agent architecture: an LLM reasoning at the application layer could call lower-layer tools via ReAct, unifying intent translation with resource allocation.
  • A testable extension is to measure the latency and energy cost of CoT/ToT on edge hardware against the millisecond physical-layer budget; if exceeded, hybrid designs that offload heavy reasoning to the cloud while keeping local fast decisions would be needed.
  • Because the survey relies on proof-of-concept numbers, a natural next step would be standardized open datasets and simulators where reasoning methods can be compared across layers under identical conditions.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. This survey proposes to fill a gap it identifies in the literature: no existing survey focuses specifically on AI reasoning techniques and their layered applications in wireless networking. It organizes reasoning methods into three categories—prompting strategies, architectural approaches, and learning paradigms—and maps them onto four task types (generation, classification, optimization, prediction). The core of the paper is a layer-by-layer tour (physical, data link, network, transport, application), with each chapter identifying challenges, presenting reasoning-based solutions, and listing typical scenarios. The final sections discuss future directions: agentic AI reasoning, data-efficient learning, efficient deployment of large models, and safe/reliable reasoning. The paper's central claim is that reasoning-based AI offers transparent, step-by-step decision making, integration of domain knowledge, and robust adaptation, and that this is a practical path toward self-driving wireless networks.

Significance. If the survey's framing is accepted, it provides a useful organizing taxonomy and a layer-by-layer research agenda that could help structure future work at the intersection of LLMs and wireless networking. The paper collects a substantial set of recent references, and the 'challenge–solution–scenario' structure makes the material accessible. It also explicitly identifies open problems such as data scarcity, deployment constraints, and safety. However, the significance is currently weakened by a mismatch between the claimed scope and the actual content, by several examples that are analogical or prospective rather than demonstrated wireless applications, and by an unresolved feasibility gap between the millisecond-scale requirements of the physical layer and the resource-intensive nature of LLM reasoning. These issues are fixable without changing the survey's fundamental design, but they need to be addressed before the paper can serve as a reliable reference.

major comments (4)
  1. [Section 9 (Conclusion) vs. Sections 3–7] The conclusion states that the survey examined 'security and cross-layer orchestration' and that representative case studies include 'UAV networks, massive MIMO, IoT random access, URLLC transport slicing, and zero-day threat detection.' The body contains no security section and no substantive cross-layer orchestration section; UAV, IoT random access, and zero-day threat detection do not appear in Sections 3–7. Only massive MIMO clearly appears. This overstates the coverage and contradicts the Introduction's promise in Section 1.3 of covering 'security issues and cross-layer collaboration.' The abstract, contributions, and conclusion must be revised to match the actual scope, or the missing topics must be added.
  2. [Section 4.1 (Data link layer)] The data link chapter's flagship ToT example is explicitly not an LLM reasoning application: the text says 'no research directly applies the ToT framework to multi-user access and scheduling problems' and instead describes RPS, a conventional scheduling algorithm. Similarly, the RAG error-control example [120] is an ASR error-correction system, not a wireless system; the text says the paradigm 'can be analogously applied.' These are presented as solutions but are actually prospective analogies. The same pattern appears in Section 6.1, where CoT/ToT for congestion control are described without a concrete wireless deployment, and in the self-supervised learning passage, which cites the GPT-4 technical report rather than a wireless SSL study. The survey should clearly separate demonstrated wireless applications from hypothetical or transferred ideas; otherwise the layer-by-layer coverage cl
  3. [Sections 3 and 8.3] The physical-layer chapter sets a strict deadline: signal processing must be completed 'within milliseconds' (Section 3). Section 8.3 concedes that state-of-the-art reasoning models are 'resource-intensive, posing challenges for wireless edge deployment.' Yet no cited experiment reports end-to-end latency, throughput, or energy for a CoT/PAL/RAG pipeline. The CoT water-filling proof-of-concept [106] reports a 50% spectral-efficiency gain but no timing; the GNN results are fast only if GNNs are counted as reasoning methods. Since the conclusion claims a 'practical path' and 'real time' operation, this unresolved latency/energy gap must be stated explicitly and the claims tempered, or measured evidence must be supplied. As written, the central thesis is a research agenda rather than a demonstrated capability at the layers where real-time constraints are most demanding.
  4. [Section 2.1 and Table 2] The taxonomy classifies GNNs and self-supervised learning as 'AI reasoning methods' alongside CoT, ToT, and RAG. This is under-justified. GNNs are neural networks for graph-structured data, and SSL is a learning paradigm; neither inherently performs step-by-step logical reasoning. By this broad definition, the boundary between 'reasoning' and conventional deep learning becomes unclear, and the novelty claim—that no prior survey focuses on reasoning in wireless—loses force. The paper should either provide explicit inclusion/exclusion criteria for what counts as reasoning or rename the category (e.g., 'structured/relational AI methods') so the central claim is not equivocal.
minor comments (5)
  1. [Table 2] The legend defines single symbols for Prompting, Architectural, Learning, and for Generation, Classification, Optimization, Prediction, but the table cells contain repeated symbols (e.g., '■ ●●', '■ ●●●●') whose meaning is never explained. Please clarify whether multiple dots indicate degree of applicability or are a formatting artifact.
  2. [Section 1.1 and Table 2] CoT is cited as [17], but reference [17] is 'Langcoop: Collaborative driving with language,' which is not the Chain-of-Thought paper. The correct foundational CoT citation appears to be [142] (Wei et al.). Please correct this citation and audit the reference list for similar mismatches.
  3. [Section 6.1] The Self-Supervised Learning subsection cites [147] as the source for SSL-based wireless adaptation, but [147] is the GPT-4 technical report. This does not support the claim about SSL in wireless networks and should be replaced with an appropriate SSL-for-networking reference.
  4. [Sections 8.2 and 9] Section 8.2 is titled 'Data-Efficient Learning and Generalization,' but the conclusion refers to 'data-efficient learning and generation.' Please align the terminology.
  5. [Various] There are several grammatical slips (e.g., 'the model can improve the reasoning...' constructions, 'inefficiently needed') and some figure captions are dense. A light copyedit would improve readability.

Circularity Check

0 steps flagged

No significant circularity: the survey's taxonomy and gap claim are independent of its inputs; self-citations are illustrative, not load-bearing.

full rationale

This paper is a survey and perspective, not a derivation with fitted parameters, equations, or statistical predictions. Its central claim is a positioning claim: prior surveys did not specifically focus on reasoning techniques and their layered applications in wireless networking. That claim is checked against Table 1 and the related-work discussion, and it does not presuppose the survey's own conclusions. The layer-by-layer "challenge-AI reasoning solution-typical scenario" framework is an organizing taxonomy, not a result derived from an input it renames. The paper does contain self-citations, including [15], [38], and [51], but they are used as supporting references for general statements or future research directions (e.g., multi-LLM collaboration, agentic reasoning, and optimization as a wireless problem class). None of these self-citations is invoked as an external theorem that forces the survey's claims, and none is a fitted parameter renamed as a prediction. Section 8.3's admission that state-of-the-art reasoning models are resource-intensive and pose challenges for edge deployment is an honest limitation and a correctness risk, not a circular step: it qualifies the central thesis rather than assuming it. No equation or result in the paper is equivalent to its own input by construction. Therefore, no significant circularity is present.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

The survey introduces no original fitted constants or physical entities. It depends on domain assumptions: that reasoning techniques transfer from NLP to wireless tasks, that cited performance numbers are accurate, and that conventional methods such as GNNs and self-supervised learning count as reasoning. These assumptions are load-bearing for the survey's taxonomy, which is why they are listed as axioms.

axioms (3)
  • domain assumption LLM reasoning techniques (CoT, ToT, RAG, ReAct, etc.) transfer from general language tasks to wireless networking tasks and produce the gains claimed in the cited papers.
    The entire layer-by-layer applicability argument depends on extrapolating these methods to wireless problems; most cited evidence is proof-of-concept or preprint, and there is no deployment validation in this survey (Sections 2 to 7).
  • ad hoc to paper GNNs and self-supervised learning qualify as AI reasoning methods.
    Table 2 and Sections 3 to 7 place GNN and SSL under reasoning paradigms without a definition of reasoning that includes them. GNNs are standard neural architectures and SSL is a training paradigm, not explicit step-by-step reasoning, so this broadens the survey's core concept in ways the text does not justify.
  • domain assumption The quantitative results repeated from cited works are accurate and representative.
    The survey uses numbers such as 50% spectral efficiency gain, 63% semantic reliability improvement, and 4250:1 compression as evidence without independent verification or error bars (Sections 3.2, 6.2). If these figures are cherry-picked or misreported, the survey's thesis loses support.

pith-pipeline@v1.3.0-alltime-deepseek · 32192 in / 13868 out tokens · 136125 ms · 2026-08-04T19:30:28.753933+00:00 · methodology

0 comments
read the original abstract

Artificial Intelligence (AI) techniques play a pivotal role in optimizing wireless communication networks. However, traditional deep learning approaches often act as closed boxes, lacking the structured reasoning abilities needed to tackle complex, multi-step decision problems. This survey provides a comprehensive review and outlook of reasoning-enabled AI in wireless communication networks, with a focus on Large Language Models (LLMs) and other advanced reasoning paradigms. In particular, LLM-based agents can combine reasoning with long-term planning, memory, tool utilization, and autonomous cross-layer control to dynamically optimize network operations with minimal human intervention. We begin by outlining the evolution of intelligent wireless networking and the limitations of conventional AI methods. We then introduce emerging AI reasoning techniques. Furthermore, we establish a classification system applicable to wireless network tasks. We also present a layer-by-layer examination for AI reasoning, covering the physical, data link, network, transport, and application layers. For each part, we identify key challenges and illustrate how AI reasoning methods can improve AI-based wireless communication performance. Finally, we discuss key research directions for AI reasoning toward future wireless communication networks. By combining insights from both communications and AI, this survey aims to chart a path for integrating reasoning techniques into the next-generation wireless networks.

Figures

Figures reproduced from arXiv: 2509.09193 by Abbas Jamalipour, Dusit Niyato, Gang Sun, Haoxiang Luo, Hongfang Yu, Jiacheng Wang, Ruichen Zhang, Shiwen Mao, Wenjiao Feng, Yanhui Bian, Yinqiu Liu, Yu Yan.

Figure 1
Figure 1. Figure 1: Structure of our survey. of reasoning in wireless networks based on the logical framework of “reasoning method￾function-wireless scenario”. • We conduct a detailed analysis of the AI reasoning methods for the physical layer, link layer, network layer, transport layer, and application layer by referring to the network hierarchy. Specifically, we propose the enabling role of AI reasoning at each level by fol… view at source ↗
Figure 2
Figure 2. Figure 2: AI reasoning methods in the physical layer. (A) The prompting strategy utilizes CoT reasoning to [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: AI reasoning methods for the data link layer. (A) The prompting strategy outlines a method employing [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: AI reasoning methods for the network layer. (A) The prompt strategy employs CoT reasoning to reflect [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: AI reasoning methods for the transport layer. (A) Prompting strategies such as CoT and ToT support [PITH_FULL_IMAGE:figures/full_fig_p021_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: AI reasoning methods for the application layer. (A) Prompting strategies, such as contextual reasoning, [PITH_FULL_IMAGE:figures/full_fig_p024_6.png] view at source ↗

discussion (0)

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

Forward citations

Cited by 2 Pith papers

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

  1. LLM-Aided A* Search in Non-Geometric Network Graphs

    cs.NI 2026-06 unverdicted novelty 5.0

    LLM-generated waypoints guided by landmark distances reduce A* node expansions by ~50% in non-geometric graphs up to 2000 nodes with only marginal path cost increase.

  2. Wireless Copilot: An AI-Powered Partner for Navigating Next-Generation Wireless Complexity

    cs.NI 2025-12 conditional novelty 5.0

    Introduces a human-in-the-loop 'Wireless Copilot' framework for 6G network management, with a LAWNets simulation showing higher intent satisfaction than LLM/RL baselines.

Reference graph

Works this paper leans on

179 extracted references · 57 linked inside Pith · cited by 2 Pith papers

  1. [1]

    Qimei Cui et al. 2025. Overview of AI and communication for 6G network: fundamentals, challenges, and future research opportunities.Science China Information Sciences68, 7 (2025), 171301

    2025

  2. [2]

    Ruijin Sun et al. 2025. A Comprehensive Survey of Knowledge-Driven Deep Learning for Intelligent Wireless Network Optimization in 6G.IEEE Communications Surveys & Tutorials(2025)

    2025

  3. [3]

    Qian Mao, Fei Hu, and Qi Hao. 2018. Deep learning for intelligent wireless networks: A comprehensive survey.IEEE Communications Surveys & Tutorials20, 4 (2018), 2595–2621

    2018

  4. [4]

    Haoxiang Luo et al. 2024. Performance analysis and comparison of nonideal wireless PBFT and RAFT consensus networks in 6G communications.IEEE Internet of Things Journal11, 6 (2024), 9752–9765

    2024

  5. [5]

    Haoxiang Luo et al . 2024. ESCM: An efficient and secure communication mechanism for UAV networks.IEEE Transactions on Network and Service Management21, 3 (2024), 3124–3139

    2024

  6. [6]

    Haoxiang Luo et al. 2025. Convergence of symbiotic communications and blockchain for sustainable and trustworthy 6G wireless networks.IEEE Wireless Communications32, 2 (2025), 18–25

    2025

  7. [7]

    Haoxiang Luo et al. 2024. Symbiotic blockchain consensus: Cognitive backscatter communications-enabled wireless blockchain consensus.IEEE/ACM Transactions on Networking(2024)

    2024

  8. [8]

    Yu Zhang, Peter Tiňo, Aleš Leonardis, and Ke Tang. 2021. A survey on neural network interpretability.IEEE Transactions on Emerging Topics in Computational Intelligence5, 5 (2021), 726–742

    2021

  9. [9]

    Xudong Wang et al. 2025. Chain-of-Thought for Large Language Model-empowered Wireless Communications.arXiv preprint arXiv:2505.22320(2025)

    Pith/arXiv arXiv 2025

  10. [10]

    Jiacheng Wang et al. 2024. Generative AI for integrated sensing and communication: Insights from the physical layer perspective.IEEE Wireless Communications31, 5 (2024), 246–255

    2024

  11. [11]

    Yuji Roh, Geon Heo, and Steven Euijong Whang. 2019. A survey on data collection for machine learning: a big data-ai integration perspective.IEEE Transactions on Knowledge and Data Engineering33, 4 (2019), 1328–1347

    2019

  12. [12]

    Christo Kurisummoottil Thomas et al. 2024. Causal reasoning: Charting a revolutionary course for next-generation AI-native wireless networks.IEEE Vehicular Technology Magazine19, 1 (2024), 16–31

    2024

  13. [13]

    Haoxiang Luo, Jian Luo, and Athanasios V Vasilakos. 2024. BC4LLM: A perspective of trusted artificial intelligence when blockchain meets large language models.Neurocomputing599 (2024), 128089

    2024

  14. [14]

    Ruixiang Tang, Yu-Neng Chuang, and Xia Hu. 2024. The science of detecting LLM-generated text.Commun. ACM67, 4 (2024), 50–59. ACM Comput. Surv., Vol. 9, No. 9, Article 35. Publication date: September 2025. 35:30 H. Luo et al

    2024

  15. [15]

    Haoxiang Luo et al. 2025. Toward Edge General Intelligence with Multiple-Large Language Model (Multi-LLM): Architecture, Trust, and Orchestration.arXiv preprint arXiv:2507.00672(2025)

    Pith/arXiv arXiv 2025

  16. [16]

    Minrui Xu et al. 2024. When large language model agents meet 6G networks: Perception, grounding, and alignment. IEEE Wireless Communications31, 6 (2024), 63–71

    2024

  17. [17]

    Xiangbo Gao et al. 2025. Langcoop: Collaborative driving with language. InProceedings of the Computer Vision and Pattern Recognition Conference. 4226–4237

    2025

  18. [18]

    Ruichen Zhang et al. 2024. Interactive AI with retrieval-augmented generation for next generation networking.IEEE Network38, 6 (2024), 414–424

    2024

  19. [19]

    Yinqiu Liu et al. 2024. Hallucination-aware Optimization for Large Language Model-empowered Communications. arXiv preprint arXiv:2412.06007(2024)

    Pith/arXiv arXiv 2024

  20. [20]

    Yashrajsinh Chudasama et al. 2025. Towards Interpretable Hybrid AI: Integrating Knowledge Graphs and Symbolic Reasoning in Medicine.IEEE Access13 (2025), 39489–39509

    2025

  21. [21]

    Jiacheng Wang et al. 2025. Generative AI based secure wireless sensing for ISAC networks.IEEE Transactions on Information Forensics and Security(2025)

    2025

  22. [22]

    Hao Zhou et al. 2025. Large language model (llm) for telecommunications: A comprehensive survey on principles, key techniques, and opportunities.IEEE Communications Surveys & Tutorials27, 3 (2025), 1955–2005

    2025

  23. [23]

    Hao Zhou et al. 2025. Large language models for wireless networks: an overview from the prompt engineering perspective.IEEE Wireless Communications(2025)

    2025

  24. [24]

    Cong T Nguyen et al. 2022. Transfer learning for wireless networks: A comprehensive survey.Proc. IEEE110, 8 (2022), 1073–1115

    2022

  25. [25]

    Muhammed Golec et al. 2025. Llm-driven apt detection for 6g wireless networks: A systematic review and taxonomy. arXiv preprint arXiv:2505.18846(2025)

    arXiv 2025

  26. [26]

    Zirui Chen, Zhaoyang Zhang, and Zhaohui Yang. 2024. Big AI models for 6G wireless networks: Opportunities, challenges, and research directions.IEEE Wireless Communications31, 5 (2024), 164–172

    2024

  27. [27]

    Sifan Long et al. 2025. A Survey on Intelligent Network Operations and Performance Optimization Based on Large Language Models.IEEE Communications Surveys & Tutorials(2025)

    2025

  28. [28]

    Gordon Owusu Boateng et al. 2025. A survey on large language models for communication, network, and service management: Application insights, challenges, and future directions.IEEE Communications Surveys & Tutorials (2025)

    2025

  29. [29]

    Fenghao Zhu et al. 2025. Wireless large AI model: Shaping the AI-native future of 6G and beyond.arXiv preprint arXiv:2504.14653(2025)

    Pith/arXiv arXiv 2025

  30. [31]

    Zhijun Chen et al. 2025. Harnessing Multiple Large Language Models: A Survey on LLM Ensemble.arXiv preprint arXiv:2502.18036(2025)

    Pith/arXiv arXiv 2025

  31. [32]

    Fei Yu et al. 2024. Natural language reasoning, a survey.Comput. Surveys56, 12 (2024), 1–39

    2024

  32. [33]

    Lauren Nicole DeLong et al. 2025. Neurosymbolic AI for reasoning over knowledge graphs: A survey.IEEE Transactions on Neural Networks and Learning Systems36, 5 (2025), 7822–7842

    2025

  33. [34]

    Yiqi Wang et al. 2024. Exploring the reasoning abilities of multimodal large language models (mllms): A comprehensive survey on emerging trends in multimodal reasoning.arXiv preprint arXiv:2401.06805(2024)

    Pith/arXiv arXiv 2024

  34. [35]

    Yunxin Li et al. 2025. Perception, reason, think, and plan: A survey on large multimodal reasoning models.arXiv preprint arXiv:2505.04921(2025)

    Pith/arXiv arXiv 2025

  35. [36]

    Avinash Patil and Aryan Jadon. 2025. Advancing reasoning in large language models: Promising methods and approaches.arXiv preprint arXiv:2502.03671(2025)

    Pith/arXiv arXiv 2025

  36. [37]

    Haoxiang Luo et al. 2025. A Trustworthy Multi-LLM Network: Challenges, Solutions, and A Use Case.arXiv preprint arXiv:2505.03196(2025)

    Pith/arXiv arXiv 2025

  37. [38]

    Haoxiang Luo et al. 2025. A Weighted Byzantine Fault Tolerance Consensus Driven Trusted Multiple Large Language Models Network.arXiv preprint arXiv:2505.05103(2025)

    Pith/arXiv arXiv 2025

  38. [39]

    Adarsh Prasad Behera et al. 2025. Towards Efficient Multi-LLM Inference: Characterization and Analysis of LLM Routing and Hierarchical Techniques.arXiv preprint arXiv:2506.06579(2025)

    Pith/arXiv arXiv 2025

  39. [40]

    Xuezhi Wang et al. 2022. Self-consistency improves chain of thought reasoning in language models.arXiv preprint arXiv:2203.11171(2022)

    Pith/arXiv arXiv 2022

  40. [41]

    Shunyu Yao et al. 2023. Tree of thoughts: Deliberate problem solving with large language models.Advances in neural information processing systems36 (2023), 11809–11822

    2023

  41. [42]

    Yao Yao et al. 2023. Beyond chain-of-thought, effective graph-of-thought reasoning in language models.arXiv preprint arXiv:2305.16582(2023). ACM Comput. Surv., Vol. 9, No. 9, Article 35. Publication date: September 2025. AI Reasoning for Wireless Communications and Networking: A Survey and Perspectives 35:31

    Pith/arXiv arXiv 2023

  42. [43]

    Guangyuan Liu et al. 2025. Adaptive Contextual Caching for Mobile Edge Large Language Model Service.arXiv preprint arXiv:2501.09383(2025)

    Pith/arXiv arXiv 2025

  43. [44]

    Azuka Chiejina et al. 2025. ACCORD: Application Context-aware Cross-layer Optimization and Resource Design for 5G/NextG Machine-centric Applications.arXiv preprint arXiv:2502.20320(2025)

    Pith/arXiv arXiv 2025

  44. [45]

    Shunyu Yao et al. 2023. React: Synergizing reasoning and acting in language models. InInternational Conference on Learning Representations (ICLR)

    2023

  45. [46]

    Luyu Gao et al. 2023. Pal: Program-aided language models. InInternational Conference on Machine Learning. PMLR, 10764–10799

    2023

  46. [47]

    Ching-Yun Ko et al. 2024. MemReasoner: A Memory-augmented LLM Architecture for Multi-hop Reasoning. InThe First Workshop on System-2 Reasoning at Scale, NeurIPS’24

    2024

  47. [48]

    Yifei Shen et al. 2022. Graph neural networks for wireless communications: From theory to practice.IEEE Transactions on Wireless Communications22, 5 (2022), 3554–3569

    2022

  48. [49]

    Jingzhi others Fang. 2025. Improving the End-to-End Efficiency of Offline Inference for Multi-LLM Applications Based on Sampling and Simulation.arXiv preprint arXiv:2503.16893(2025)

    Pith/arXiv arXiv 2025

  49. [50]

    Christo Kurisummoottil Thomas and Walid Saad. 2023. Neuro-symbolic causal reasoning meets signaling game for emergent semantic communications.IEEE Transactions on Wireless Communications23, 5 (2023), 4546–4563

    2023

  50. [51]

    Yinqiu Liu et al. 2025. LAMeTA: Intent-Aware Agentic Network Optimization via a Large AI Model-Empowered Two-Stage Approach.arXiv preprint arXiv:2505.12247(2025)

    Pith/arXiv arXiv 2025

  51. [52]

    Aman Madaan et al . 2023. Self-refine: Iterative refinement with self-feedback.Advances in Neural Information Processing Systems36 (2023), 46534–46594

    2023

  52. [53]

    Xiang Li et al. 2024. MoDE-CoTD: Chain-of-Thought Distillation for Complex Reasoning Tasks with Mixture of Decoupled LoRA-Experts. InProceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024). 11475–11485

    2024

  53. [54]

    Yang Yue et al. 2025. Does reinforcement learning really incentivize reasoning capacity in llms beyond the base model?arXiv preprint arXiv:2504.13837(2025)

    Pith/arXiv arXiv 2025

  54. [55]

    Eric Zelikman et al. 2022. Star: Bootstrapping reasoning with reasoning, 2022.URL https://arxiv. org/abs/2203.14465 2203 (2022)

    Pith/arXiv arXiv 2022

  55. [56]

    Dimitrios Sikeridis et al. 2025. Pickllm: Context-aware rl-assisted large language model routing. InInternational Workshop on AI for Transportation. Springer, 227–239

    2025

  56. [57]

    Ruixin Hong, Xinyu Pang, and Changshui Zhang. 2024. Advances in reasoning by prompting large language models: A survey.Cybernetics and Intelligence(2024)

    2024

  57. [58]

    Jiangnan Li et al. 2024. Think out Loud: Emotion Deducing Explanation in Dialogues.arXiv preprint arXiv:2406.04758 (2024)

    Pith/arXiv arXiv 2024

  58. [59]

    Jia Li et al . 2025. Structured chain-of-thought prompting for code generation.ACM Transactions on Software Engineering and Methodology34, 2 (2025), 1–23

    2025

  59. [60]

    Takeshi Kojima et al. 2022. Large language models are zero-shot reasoners.Advances in neural information processing systems35 (2022), 22199–22213

    2022

  60. [61]

    Seungone Kim et al. 2023. The cot collection: Improving zero-shot and few-shot learning of language models via chain-of-thought fine-tuning.arXiv preprint arXiv:2305.14045(2023)

    Pith/arXiv arXiv 2023

  61. [62]

    Alan Boyle et al. 2024. iToT: An Interactive System for Customized Tree-of-Thought Generation.arXiv preprint arXiv:2409.00413(2024)

    Pith/arXiv arXiv 2024

  62. [63]

    Jiankai Sun et al. 2025. A survey of reasoning with foundation models: Concepts, methodologies, and outlook.Comput. Surveys57, 11 (2025), 1–43

    2025

  63. [64]

    Jianmin Ye et al. 2025. ChatModel: Automating Reference Model Design and Verification with LLMs.arXiv preprint arXiv:2506.15066(2025)

    Pith/arXiv arXiv 2025

  64. [65]

    Zhengyuan Yang et al. 2023. Mm-react: Prompting chatgpt for multimodal reasoning and action.arXiv preprint arXiv:2303.11381(2023)

    Pith/arXiv arXiv 2023

  65. [66]

    Renat Aksitov et al. 2023. Rest meets react: Self-improvement for multi-step reasoning llm agent.arXiv preprint arXiv:2312.10003(2023)

    Pith/arXiv arXiv 2023

  66. [67]

    Dayu Yang et al. 2025. Code to think, think to code: A survey on code-enhanced reasoning and reasoning-driven code intelligence in llms.arXiv preprint arXiv:2502.19411(2025)

    Pith/arXiv arXiv 2025

  67. [68]

    Zhuocheng Shen. 2024. Llm with tools: A survey.arXiv preprint arXiv:2409.18807(2024)

    Pith/arXiv arXiv 2024

  68. [69]

    Maaike De Boer et al . 2025. Design Patterns for LLM-based Neuro-Symbolic Systems.Neurosymbolic Artificial Intelligence(2025)

    2025

  69. [70]

    Mohd Ariful Haque et al. 2025. Advanced Tool Learning and Selection System (ATLASS): A Closed-Loop Framework Using LLM.arXiv preprint arXiv:2503.10071(2025). ACM Comput. Surv., Vol. 9, No. 9, Article 35. Publication date: September 2025. 35:32 H. Luo et al

    arXiv 2025

  70. [71]

    Elias Lumer et al. 2024. Toolshed: Scale tool-equipped agents with advanced rag-tool fusion and tool knowledge bases.arXiv preprint arXiv:2410.14594(2024)

    Pith/arXiv arXiv 2024

  71. [72]

    Yingzhe Peng et al. 2025. Lmm-r1: Empowering 3b lmms with strong reasoning abilities through two-stage rule-based rl.arXiv preprint arXiv:2503.07536(2025)

    Pith/arXiv arXiv 2025

  72. [73]

    Wenda Xu et al. 2024. Pride and prejudice: LLM amplifies self-bias in self-refinement.arXiv preprint arXiv:2402.11436 (2024)

    Pith/arXiv arXiv 2024

  73. [74]

    Zhengyuan Yang et al. 2024. Idea2img: Iterative self-refinement with gpt-4v for automatic image design and generation. InEuropean Conference on Computer Vision. Springer, 167–184

    2024

  74. [75]

    Yuan Sui et al. 2025. Meta-reasoner: Dynamic guidance for optimized inference-time reasoning in large language models.arXiv preprint arXiv:2502.19918(2025)

    Pith/arXiv arXiv 2025

  75. [76]

    Xunyu Zhu et al. 2024. Distilling mathematical reasoning capabilities into small language models.Neural Networks 179 (2024), 106594

    2024

  76. [77]

    Chiwei Zhu et al. 2025. Rationales Are Not Silver Bullets: Measuring the Impact of Rationales on Model Performance and Reliability.arXiv preprint arXiv:2505.24147(2025)

    Pith/arXiv arXiv 2025

  77. [78]

    Ruichen Zhang et al. 2025. Embodied AI-Enhanced Vehicular Networks: An Integrated Vision Language Models and Reinforcement Learning Method.IEEE Transactions on Mobile Computing(2025)

    2025

  78. [79]

    Robert R Underwood et al . 2023. Libpressio-predict: Flexible and fast infrastructure for inferring compression performance. InProceedings of the SC’23 Workshops of the International Conference on High Performance Computing, Network, Storage, and Analysis. 272–280

    2023

  79. [80]

    Yue Zheng et al. 2025. A review on edge large language models: Design, execution, and applications.Comput. Surveys 57, 8 (2025), 1–35

    2025

  80. [81]

    Fahime Khoramnejad and Ekram Hossain. 2025. Generative AI for the optimization of next-generation wireless networks: Basics, state-of-the-art, and open challenges.IEEE Communications Surveys & Tutorials(2025)

    2025

Showing first 80 references.