{"id":"3dd17711-7b66-47ae-9220-f0f7454870f6","arxiv_id":"2506.09512","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":2.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A brief survey of AI/ML methods for 6G-V2X communications, covering DL, RL, generative learning, and federated learning across four application areas.","lead":"This paper surveys recent uses of artificial intelligence in 6G vehicle-to-everything (V2X) communication, grouping methods into deep learning, reinforcement learning, generative learning, and federated learning. It is a short overview meant to help researchers and engineers get a quick map of the field.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The survey's 'comprehensive' claim is unsupported by a reproducible selection methodology; a systematic search would likely reveal substantial omitted work, undermining the central claim.","rationale":"The reader's weakest_assumption identifies exactly the same load-bearing concern: the survey's comprehensiveness and accuracy rest on an undocumented, non-reproducible selection of papers. My stress-test confirms this and adds specificity: the article's own body and Table I show a small set of references per category (1–3 per row), several pre-2023, and incomplete citations, making it impossible for a reader to trust the 'comprehensive' label. The concrete test I propose would settle the concern by empirically measuring coverage against a systematic search and verifying summary accuracy. Because the reader already set the verdict to CONDITIONAL, and my analysis corroborates rather than overturns that judgment, keeping CONDITIONAL is appropriate: the paper should be accepted only if the authors add a clear methodology section and demonstrate that their selection is representative, or revise the claim to 'a selective overview'. Therefore verdict_should_be is UNCHANGED relative to the reader's decision.","tokens_in":11168,"tokens_out":2643,"duration_ms":28923,"concrete_test":"Perform an independent systematic literature search in IEEE Xplore, Scopus, and Web of Science for the period 2023–2025 using a query such as TITLE-ABS-KEY(('6G' AND 'V2X') AND ('machine learning' OR 'deep learning' OR 'reinforcement learning' OR 'generative' OR 'federated learning')) limited to journal and conference papers. Retrieve the full result set, deduplicate it, and compare it against the list of primary papers cited in the survey (excluding the survey's own background/vision references). If the search returns more than, say, 50 relevant papers, and more than 50% of them are not cited or discussed in any form, then the 'comprehensive' claim is empirically false.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim, stated in the abstract and Section I, is that this survey 'comprehensively reviews recent advances in AI and ML models applied to 6G-V2X communication' with emphasis on 2023–2025. The load-bearing premise is that the selected papers are representative and that the coverage is broad enough to justify the term 'comprehensive'. This premise is not established. Section I states only that the selection 'focuses on models that demonstrate high performance, adaptability, and effectiveness' but gives no inclusion/exclusion criteria, no search strategy, no databases, and no time-window definition beyond 'past two years'. The body then discusses only about 20 primary papers, organized into four application areas, with no comparison of methods, no quantitative synthesis, and no discussion of how the presented papers were selected from the larger literature. For example, Table I lists only 14 references across all four application categories; [38], [41], and [42] are pre-2023, so the '2023–2025' emphasis is not consistently honored. Furthermore, some references are incomplete or lack author lists (e.g., [47] 'Asha et al.' with no venue, [52] 'Hangdong et al.' with no authors), which makes it impossible to verify that the summaries are accurate. If the survey omits a significant fraction of relevant recent work without justification, then the 'comprehensive review' claim fails, and the paper provides only a shallow, non-reproducible overview rather than the systematic survey it advertises. This is the single most load-bearing concern because the entire value of a survey rests on its completeness and reliability.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript is a survey of recent AI/ML techniques applied to 6G-V2X communication. After introducing the broader AI/ML landscape, it organizes the literature into four model families (ML, DL, generative learning, and federated learning) and four application areas (resource management, mmWave/THz beamforming, intelligent traffic management, and security management). The paper's central claim, stated in the abstract and Section I, is that it 'comprehensively reviews' recent advances with emphasis on 2023–2025, and it concludes with a discussion of challenges and future directions. There are no experiments; the contribution is intended as an organized overview of selected recent papers.","tokens_in":11451,"tokens_out":5145,"duration_ms":53591,"significance":"If the survey delivered on its claim, it would be a useful entry point for researchers and engineers seeking an overview of AI/ML for 6G-V2X, particularly because the topic is timely and the paper touches on emerging directions such as LLM integration and federated learning. The paper identifies several relevant recent papers and gives plausible summaries of their contributions, and the challenges section correctly names non-IID data, privacy, latency, and edge resource constraints. However, the survey's significance is currently limited by its non-reproducible selection of papers, the small number of primary works covered, and the absence of comparative or critical synthesis. The abstract's emphasis on generative learning is also not matched by the application sections, which discuss only one LLM-based application in detail.","major_comments":[{"comment":"The abstract and Section I claim a 'comprehensive' review of AI/ML for 6G-V2X with emphasis on 2023–2025, but the paper provides no reproducible selection methodology: there is no search strategy, no list of databases, no inclusion/exclusion criteria, and no definition of the 'past two years' window. Section I states only that the selection focuses on models that 'demonstrate high performance, adaptability, and effectiveness' without saying how these qualities were assessed. Because the central value of a survey depends on representative coverage, this omission is load-bearing rather than cosmetic.","section":"I (Introduction)"},{"comment":"The claim of emphasis on 2023–2025 is not consistently honored. Table I includes [38] (Prathiba et al., IEEE Transactions on Network Science and Engineering, 2021) and [41] (Moon et al., Journal of Communications and Networks, 2020) as representative entries for resource management and beamforming, and Section III.A also relies on [33] from 2017. If older works are included as background, the survey should explicitly say so; as written, the temporal claim in the abstract and Section I is contradicted by the table's own contents.","section":"Table I and Sections III.A–III.D"},{"comment":"Several cited works are incomplete to the point of being unverifiable: [31] has no author list, [47] has no authors and no venue, [52] has no authors, and [53] has no authors or venue. This matters because Sections II.D and III.D rely on these references for specific factual claims, including the robustness properties of decentralized federated learning and the FedVPS scheme. Without full bibliographic data, a reader cannot check whether the summaries are accurate or whether the cited work actually appears in the scholarly record.","section":"References [25], [31], [47], [52], [53]"},{"comment":"Section III is an annotated list of roughly twenty papers rather than the 'systematic summary' promised in the abstract. Each subsection describes selected papers individually but does not compare their methods, evaluation setups, or reported gains, nor does it explain how the chosen papers relate to the wider literature. In addition, the abstract highlights generative learning as a particularly promising direction, yet the application sections contain no detailed treatment of GANs, VAEs, or diffusion models for 6G-V2X, and the only generative method discussed at the application level is the LLM-based resource allocation in [36]. This mismatch further weakens the claim of a comprehensive and balanced review.","section":"III (AI for 6G-V2X Applications)"}],"minor_comments":[{"comment":"There are repeated typographical and formatting errors, including 'CA Vs' in the abstract, 'COMMUNCATION' in the Section II heading, 'Artifical' in reference [13], 'V AEs' in Section II.C, and 'disruption,s' in Section IV. These should be corrected in a thorough copyedit.","section":"Throughout"},{"comment":"References [51] and [54] are the same paper (Osorio et al., 'Towards 6G-enabled Internet of Vehicles: Security and Privacy,' IEEE Open Journal of the Communications Society, 2022) but are cited as if they were distinct sources in the security discussion and in the challenges section.","section":"Section III.D and references [51], [54]"},{"comment":"The in-text attribution 'Asha et al. [47]' cannot be verified against the reference list, since [47] contains no author names. Similarly, '[52] Hangdong et al.' has no author list in the bibliography; these attributions should be reconciled with complete references.","section":"Section III.D"},{"comment":"The sentence 'In this paper [34], this paper introduces a GNN and DRL method' is grammatically redundant and should be rewritten, for example as '[34] introduces a GNN and DRL method.'","section":"Section III.A"},{"comment":"The introduction mentions that earlier surveys are outdated and that certain emerging 6G-V2X applications are not covered, but it does not specify which emerging applications are newly covered here; a brief comparison with prior surveys would help clarify the claimed contribution.","section":"Section I"}],"recommendation":"major_revision","confidential_remarks":"The core issue is the gap between the claim of comprehensiveness and the evidence presented. This is fixable within the scope of a survey paper: a revision that adds a methodology section, expands or justifies the set of covered papers, resolves the temporal inconsistency in Table I, and either substantiates or softens the generative-learning emphasis would make the manuscript publishable. I do not see a deeper circularity or integrity problem; the inaccuracies appear to be presentation and coverage issues rather than misrepresentation of the summarized results."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a competent but shallow survey. If you want a 15-minute orientation to AI/ML in 6G-V2X, it works. If you want a comprehensive, up-to-date map of the field, it doesn't, and the title oversells it.\n\nWhat's good: the authors organize recent literature into four AI families (DL, RL, generative learning, federated learning) and map them onto four application areas (resource management, beamforming, traffic management, security). The generative-learning emphasis, especially LLMs for V2X, is timely and not common in earlier surveys. The individual paper summaries are mostly faithful to the cited abstracts, and the challenges section, though generic, hits the right issues. The self-citation [45] is incidental, not load-bearing.\n\nWhere it falls short: the central claim, stated in the abstract and Introduction, is that this is a comprehensive review of 2023–2025 work. There is no methodology to back that up: no search strategy, no inclusion/exclusion criteria, no database list. The body only discusses about twenty primary papers, and Table I lists fourteen references, some from before 2023. That may be fine for a 'representative models' review, but it is not comprehensive. The reference list has real problems: [31], [47], [52], and [53] lack author lists or full titles, which makes verification hard. The paper also leans heavily on earlier surveys [10] and [11] and doesn't offer a systematic comparison or quantitative synthesis of the included works. In short, the paper's value is as a readable entry point, not as a definitive survey.\n\nThe comprehensiveness claim is load-bearing for a survey, and it isn't supported. That said, this is fixable: retitle and rescope the paper as a review of representative AI/ML models, add a short methodology paragraph, and complete the references. With those revisions, it would be a reasonable paper.\n\nThis is for newcomers—students or engineers just entering 6G-V2X who want a quick map. Experts won't find new information. I'd send it to peer review because it's a coherent, potentially useful survey in a hot area, but I'd make major revision the likely outcome.\n\nRecommendation: engage with it, but require the rescoping and reference fixes.","headline":"A readable but mis-titled survey: the 'comprehensive' claim isn't supported, yet it's a useful entry point for newcomers to AI/ML in 6G-V2X.","tokens_in":11974,"tokens_out":3133,"would_cite":false,"duration_ms":32440,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This survey claims that AI and machine learning, especially generative learning, have advanced enough to materially improve 6G Vehicle-to-Everything (V2X) communication, and that the field now needs—and receives—a systematic up-to-date…","keywords":["6G","V2X","machine learning","deep learning","reinforcement learning","generative learning","federated learning","intelligent transportation systems"],"falsifier":"A reader could compile a list of 2023-2025 studies on AI for 6G-V2X using a defined keyword query across standard scholarly databases; if many relevant works meeting the same criteria are absent from this survey, or if checking the cited papers shows their reported results are misstated, the claim of a comprehensive, reliable review would be falsified.","tokens_in":10988,"feed_emoji":"🚗","tokens_out":6072,"duration_ms":58033,"temperature":0.7,"pith_summary":"This survey claims that the fast-moving field of AI for 6G Vehicle-to-Everything (V2X) communication had, until now, lacked a systematic up-to-date summary, and it sets out to provide one. It reviews work from roughly 2023 to 2025, organizing the landscape into deep learning, reinforcement learning, generative learning (including large language models), and federated learning. It then shows how each family has been applied to intelligent resource allocation, beamforming in millimeter-wave and terahertz bands, intelligent traffic management, and security. A sympathetic reader would take the paper's contribution to be a reliable map of the current state of the art and of the open challenges, especially the emerging role of generative AI.","feed_headline":"Survey maps AI's growing role in 6G vehicle-to-everything","feed_subtitle":"Covers deep, reinforcement, generative, and federated learning across four 6G-V2X use cases.","key_machinery":"The organizing device is a four-by-four taxonomy: four AI and machine-learning model families—Deep Learning, Reinforcement Learning, Generative Learning (GANs, VAEs, LLMs), and Federated Learning—mapped onto four 6G-V2X application areas—intelligent resource management, AI-powered beamforming in mmWave and THz bands, intelligent traffic management, and security management. The taxonomy is summarized in Table I, which associates representative references with each cell. The survey uses this grid to structure its review and to argue that each model family has a distinct role: deep reinforcement learning for dynamic decision-making, deep learning for channel estimation, generative models for producing realistic content and safety messages, and federated learning for privacy-preserving distributed training.","core_discovery":"The paper's central claim is that recent advances in AI and machine learning—particularly generative learning—have progressed to the point where they can materially improve the performance, adaptability, and intelligence of 6G-V2X systems, and that this progress is best understood through a taxonomy of four model families applied to four application domains. The survey assembles representative studies from 2023 to 2025 and reports their stated gains, for example a 20.69% improvement in computation offloading and resource allocation from a federated learning framework, a 94.2% privacy protection score, and up to 97% intrusion-detection accuracy. It concludes that integrating AI with 6G-V2X is the direction of future mobility, while identifying real-time processing, data privacy, non-IID data, edge resource limits, scalability, and missing standards as the main obstacles.","pith_inferences":["Because the survey does not describe a systematic search protocol or inclusion criteria, its comprehensiveness is a claim about selected representative studies rather than a guarantee of full coverage; a reader should treat the map as an orientation, not a census.","If generative learning is indeed the fastest-moving branch, one testable prediction is that the share of 6G-V2X papers using LLMs or other generative models will grow faster than the share using classical deep learning or reinforcement learning over the next two to three years.","The security results the survey cites are mostly simulation-based; a natural extension would be to benchmark the same AI models on a common real-world V2X security dataset to see whether the reported accuracy and privacy scores transfer.","The survey's taxonomy could be operationalized: a future systematic review could tag each paper by model family and application area, producing a quantitative map that would test whether the four-by-four grid adequately covers the field."],"forward_implications":["If the survey's map is accurate, a researcher entering 6G-V2X can use it to identify the dominant AI approach for a given application area and the representative papers behind it.","The survey's emphasis on 2023 to 2025 suggests that deep reinforcement learning and federated learning have become the default tools for resource allocation and beamforming, rather than classical optimization alone.","Generative learning and large language models appear as the newest and least settled direction; the survey implies they will increasingly handle message generation, scenario simulation, and natural-language interaction in vehicular networks.","The listed challenges—latency, privacy, non-IID data, edge resource limits, and missing standards—define the near-term engineering agenda for AI-driven 6G-V2X deployment."],"supporting_citations":[{"why":"Supplies the 6G-V2X technology background and the framing of machine learning as instrumental for vehicular networking.","marker":"[1]"},{"why":"An earlier AI-for-V2X survey that this paper positions itself as updating and extending.","marker":"[10]"},{"why":"The other earlier AI/ML-for-V2X survey whose coverage this paper argues is partly outdated.","marker":"[11]"},{"why":"A survey of deep reinforcement learning for autonomous-vehicle motion planning, used as representative DL/DRL background.","marker":"[20]"},{"why":"Links generative AI to vehicular networks, grounding the paper's claim that GL is an emerging and promising direction.","marker":"[23]"},{"why":"Surveys LLMs and AI agents in V2X, supporting the treatment of LLMs as a key generative-learning application.","marker":"[29]"},{"why":"A survey of federated learning in intelligent transportation systems that anchors the FL section.","marker":"[32]"},{"why":"A federated-learning resource management framework whose reported 20.69% improvement is cited as evidence for FL's effectiveness in 6G-V2X.","marker":"[38]"},{"why":"A deep-learning channel estimation and tracking method cited as evidence for DL in mmWave V2X beamforming.","marker":"[41]"},{"why":"A DRL-based traffic signal control study cited as evidence for AI in intelligent traffic management.","marker":"[46]"}],"fun_headline_variants":["Generative AI emerges as key for 6G V2X","Survey: Four AI families optimize 6G V2X","AI drives 6G V2X improvements, survey finds","Deep, RL, generative, federated: AI in 6G V2X"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The survey's usefulness rests on the assumption that the papers it selected are representative of the 2023-2025 6G-V2X AI landscape and that its summaries of them are faithful, since no systematic search or inclusion criteria are described.","fun_headline_variants_meta":{"raw":{"variants":["Generative AI emerges as key for 6G V2X","Survey: Four AI families optimize 6G V2X","AI drives 6G V2X improvements, survey finds","Deep, RL, generative, federated: AI in 6G V2X"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000456,"raw_usage":{"total_tokens":2328,"prompt_tokens":1025,"completion_tokens":1303,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":641,"completion_tokens_details":{"reasoning_tokens":1226}},"tokens_in":641,"tokens_out":1303,"duration_ms":12319,"temperature":1.0,"reasoning_tokens":1226,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T04:45:23.829718+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A reader could compile a list of 2023-2025 studies on AI for 6G-V2X using a defined keyword query across standard scholarly databases; if many relevant works meeting the same criteria are absent from this survey, or if checking the cited papers shows their reported results are misstated, the claim of a comprehensive, reliable review would be falsified.","supporting_citations":[{"cited_title":"6g for vehicle-to-everything (v2x) communications: Enabling technologies, challenges, and oppor- tunities,","cited_arxiv_id":null,"evidence_quote":"Supplies the 6G-V2X technology background and the framing of machine learning as instrumental for vehicular networking."},{"cited_title":"Artificial intelli- gence for vehicle-to-everything: A survey,","cited_arxiv_id":null,"evidence_quote":"An earlier AI-for-V2X survey that this paper positions itself as updating and extending."},{"cited_title":"Artificial intelligence and machine learning as key enablers for v2x communications: A comprehensive survey,","cited_arxiv_id":null,"evidence_quote":"The other earlier AI/ML-for-V2X survey whose coverage this paper argues is partly outdated."},{"cited_title":"Survey of deep reinforcement learning for motion planning of autonomous vehicles,","cited_arxiv_id":null,"evidence_quote":"A survey of deep reinforcement learning for autonomous-vehicle motion planning, used as representative DL/DRL background."},{"cited_title":"Generative ai-enabled vehicular networks: Fundamentals, framework, and case study,","cited_arxiv_id":null,"evidence_quote":"Links generative AI to vehicular networks, grounding the paper's claim that GL is an emerging and promising direction."},{"cited_title":"Survey of llms and ai agents in v2x: Simulation, analysis & architectures,","cited_arxiv_id":null,"evidence_quote":"Surveys LLMs and AI agents in V2X, supporting the treatment of LLMs as a key generative-learning application."},{"cited_title":"Federated learning empowered computation offloading and resource management in 6g-v2x,","cited_arxiv_id":null,"evidence_quote":"A federated-learning resource management framework whose reported 20.69% improvement is cited as evidence for FL's effectiveness in 6G-V2X."},{"cited_title":"Deep learning-based channel esti- mation and tracking for millimeter-wave vehicular communications,","cited_arxiv_id":null,"evidence_quote":"A deep-learning channel estimation and tracking method cited as evidence for DL in mmWave V2X beamforming."},{"cited_title":"Intelligent road network management supported by 6g and deep reinforcement learning,","cited_arxiv_id":null,"evidence_quote":"A DRL-based traffic signal control study cited as evidence for AI in intelligent traffic management."}],"review_version":1}