Pith. sign in

REVIEW 4 major objections 6 minor 28 references

Blockchain-Linked Auditable Decision Management for Telecom/IoT Fraud-Control Requests

T0 review · 4 major / 6 minor · reviewed 2026-07-13 · grok-4.5

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

Pith's one-line read QLoRA-tuned LLM risk scoring approaches but does not beat a lower-cost centralized ensemble for telecom/IoT fraud-control request decisions under controlled synthetic deployment replay.

desk verdict Honest systems comparison under a shared request/policy/audit substrate: QLoRA becomes usable but does not beat a cheaper ensemble; the ranking is real for their generator and carefully scoped, not field proof. read the letter →

arxiv 2607.09259 v1 pith:IYUNKMN6 submitted 2026-07-10 cs.CR cs.AI

Saviz Changizi , Nasibeh Mohammadzadeh , Mohammad Shojafar , Rahim Tafazolli This is my paper
classification cs.CRcs.AI
keywords TelecomfraudcontrolIoTsecurityauditabledecisionmanagementblockchainauditfederatedlearninglargelanguagemodelsQLoRAdeploymentreplay
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Telecom fraud work usually stops at scoring suspicious records. This paper instead treats fraud control as a full request-management problem: each synthetic service-usage record becomes a managed request that must be approved or blocked, with a durable audit trail. Explicit out-of-boundary cases are blocked by a fixed hard-fraud gate; everything else is scored by one of three risk sources—centralized machine learning, federated meta-learning, or an LLM family—and resolved through the same five-state policy and two-zone refinement before a local Ethereum-compatible layer logs the lifecycle. The main empirical claim is that fine-tuning the LLM branch with QLoRA makes it far more usable than zero-shot prompting, cutting legitimate false positives sharply and raising soft-fraud recall, yet it still mainly approaches rather than surpasses a cheaper centralized ensemble. On a 100,000-record drifted replay the legitimate false-positive rates of both the ensemble and QLoRA rise well above the validation 10 percent cap, so the authors present the numbers as controlled drift-replay evidence, not live-operator validation.

What carries the argument

Shared deployment-stage decision protocol: deterministic hard-fraud (out-of-boundary) gate first, then configuration-specific non-hard risk score mapped into NO / MAYBE_LOW / MAYBE_HIGH / YES, two-zone refinement of the ambiguous states, and a common Ethereum-compatible lifecycle that only records already-formed APPROVE/BLOCK actions.

What would settle it

Re-run the identical frozen policy and audit pipeline on operator-supported or otherwise independently labeled telecom/IoT request streams; if the QLoRA branch no longer approaches the centralized ensemble on legitimate false-positive rate and soft-fraud recall, or if blockchain telemetry ceases to be explained solely by off-chain decision profiles, the central claim fails.

Watch

Extended reading notes

Core claim

Under a shared hard-fraud gate, five-state policy, two-zone refinement, and blockchain audit path, QLoRA sequence-classification risk scores become operationally usable and close the gap to a lower-cost centralized ensemble, while zero-shot LLM artifacts remain too aggressive; blockchain gas, latency, and throughput differences track only the submitted off-chain decision profiles.

Load-bearing premise

The synthetic training and 100,000-record deployment corpora—with their eight-type soft/hard fraud taxonomy, design-specific operating bounds, and injected moderate drift—are faithful enough proxies that the ranking of risk sources and the size of the legitimate false-positive gap would hold for real operator traffic.

Editorial extensions

If this is right

  • Request-level policy resolution and lifecycle audit become first-class evaluation objects, not afterthoughts to detector accuracy.
  • Zero-shot structured LLM probabilities should not be used as direct operational risk scores without supervised adaptation.
  • QLoRA-tuned sequence classification can serve as a usable non-hard risk backbone, but cost-conscious operators can still prefer a calibrated centralized ensemble.
  • Blockchain audit cost and latency are predictable from the off-chain decision profile alone; changing consensus-inspired validator assignment does not alter fraud outcomes.
  • Validation-time false-positive caps do not guarantee the same operating point under even moderate contextual drift.

Reading between the lines

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

  • Any production deployment would need continuous drift monitoring and threshold recalibration; the observed FPR inflation under controlled drift already signals that frozen validation policies will age quickly.
  • The hard/soft split plus two-zone refinement is a reusable pattern for other high-stakes access-control or spectrum-access decisions that mix deterministic rules with probabilistic scores.
  • Because blockchain only records decisions, operators can swap risk sources without rewriting the smart-contract path—an architectural separation that may matter more for compliance than for raw accuracy.
  • Soft-fraud single-component misses remain the residual failure mode; targeted features for spoofed location and bandwidth spikes look like the highest-leverage next engineering step.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper reframes telecom/IoT fraud control as request-level auditable decision management rather than detector-only classification. In a controlled synthetic setting, each deployment record is mapped to a managed request; a deterministic hard-fraud (OOB) gate blocks explicit boundary violations, while non-hard requests are scored by centralized ML (M1/AVG6), federated meta-learning over fixed base detectors (M2), or LLM-family sources (M3-Base zero-shot vs M3-QLoRA sequence classification). Shared five-state policy resolution with two-zone refinement and a local Ethereum-compatible audit layer then produce APPROVE/BLOCK actions and lifecycle telemetry. Using separate synthetic training data and a 100,000-record deployment-replay corpus, the main empirical claim is that M3-QLoRA is far more usable than zero-shot M3-Base but mainly approaches rather than outperforms lower-cost M1 (validation: M1 soft-fraud recall 0.8341, legit. FPR 0.0890; labeled deployment: M1 FPR 0.1646 / soft recall 0.8300 vs M3-QLoRA 0.1801 / 0.8240; M3-Base legit. FPR 0.3915). Blockchain gas/latency/throughput differences track submitted off-chain decision profiles, not fraud logic.

Significance. If the controlled ranking and framework hold under the stated synthetic protocol, the paper offers a useful systems contribution: a unified request substrate that fairly compares heterogeneous risk sources under fixed policy and audit, with explicit hard/soft separation, leakage-aware split-aware preprocessing, and an off-chain/on-chain separation that treats blockchain as lifecycle audit rather than a detector. The careful scoping to drift-replay (not field validation) and the quantitative demonstration that QLoRA sequence classification largely repairs zero-shot operational aggressiveness are valuable for the telecom/IoT security and auditable decision-management communities. Strengths include schema-aligned corpora, frozen validation-selected policy under an explicit FPR cap, diagnostic soft-fraud complexity/component breakdowns (Tables VII–VIII, Fig. 4), and transparent blockchain telemetry. The result is incremental relative to detector-centric fraud work and the authors’ prior conference hybrid study, but the unified deployment-stage comparison is a legitimate journal-level systems result if limitations and cost claims are tightened.

major comments (4)
  1. [Abstract / §VIII / Table IX] Abstract and §VIII assert that M3-QLoRA “mainly approaches, rather than outperforms, a lower-cost centralized ensemble,” but the manuscript never quantifies risk-source cost (training compute, inference latency, memory, or energy) for M1 vs M3-QLoRA/M3-Base. Only blockchain lifecycle gas/latency/throughput are reported (Table IX), and those are driven by review/finalization volume. For a systems claim that hinges on “lower-cost,” add direct off-chain cost/latency measurements (or remove/qualify the cost language).
  2. [§V-A, Table IV, §V-E, Table VIII, §VII] §V-A, Table IV and §V-E couple soft fraud (in-range deviations of latency, bandwidth, power, signal, duration, location, band, off-hour activity) to the same runtime variables used by the deterministic OOB hard gate and by the tabular/sequence feature schema. Residual errors concentrate in single-component soft fraud while multi-component cases are recovered more strongly by M1/M2 (Table VIII)—the pattern expected if the generator’s coordinated numeric deviations favor calibrated tabular ensembles. This does not invalidate the within-corpus ranking, but it is load-bearing for interpreting “approaches rather than outperforms” as a transferable risk-source conclusion. Strengthen §VII with this specific generator-alignment risk and, if feasible, an ablation or sensitivity check (e.g., soft-fraud cues less aligned with the OOB envelope, or non-tabular co-fraud structure).
  3. [§III (R_{0.10,1.00}), Table VI, Table VII] Validation operating selection uses R_{0.10,1.00} (legit. FPR ceiling 0.10), yet labeled deployment legit. FPR rises to 0.1646 (M1), 0.1927 (M2), 0.1801 (M3-QLoRA), and 0.3915 (M3-Base) (Table VII). The paper correctly attributes this to controlled drift and different fraud mix, but the frozen five-state boundaries and refinement thresholds (Table VI) are then no longer operating under the advertised cap. Either report a deployment-side recalibrated operating point under the same FPR rule for fair comparison, or quantify how much of the M1–M3-QLoRA gap is threshold-shift vs score-quality under drift; otherwise the operational meaning of the shared policy is unclear.
  4. [Table VII / §VI-B] The M1 vs M3-QLoRA ranking (soft recall ~0.83 vs ~0.82; deployment FPR 0.1646 vs 0.1801) is reported as point estimates from a single frozen pipeline with no uncertainty (bootstrap CIs, multi-seed QLoRA runs, or sensitivity to boundary selection). Given how close the deployment soft-recall numbers are and how large the FPR drift is, add uncertainty quantification or multi-run stability so the “approaches rather than outperforms” claim is statistically supported rather than descriptive.
minor comments (6)
  1. [Abstract, §IV-A, §V-G] M2 is a federated meta-learner over fixed base-detector probabilities (7 device-type clients, 12 rounds), not end-to-end federated training of the detectors. The text states this, but the abstract/intro phrasing “federated meta-learning (M2)” can still be misread as full FL; tighten terminology consistently.
  2. [Fig. 2, Fig. 5] Fig. 2 and Fig. 5 are informative but dense; ensure state counts and percentages are legible in grayscale and that HARD_FRAUD (fixed at 4,500) is visually distinguished from configuration-dependent non-hard states.
  3. [Table I / §II] Table I is a useful conceptual positioning table; a short sentence on why numerical head-to-head comparison with prior detector papers is intentionally omitted would help readers who expect benchmark-style rows.
  4. [Algorithm 1, §III] Algorithm 1 and Table II are clear; define src_{i,k} and the abstract functions (RISKSIGNAL, ZONEREFINER, etc.) once in the main text near the algorithm for readers who skip the appendix-style detail.
  5. [Throughout] Minor consistency: “A VG6” / “AVG6” spacing and “V oIP” / “VoIP” appear with odd spaces in the compiled text; clean for production.
  6. [§V-J, Table X, §VII-B] Blockchain modes (PoS/PoW/PoA/DPoS/PBFT) are correctly labeled as validator-assignment policies on Ganache, not native consensus; keep that caveat adjacent to Table X so telemetry is not over-read.

Circularity Check

1 steps flagged · score 2.0 of 10

Minor architectural tautology on blockchain telemetry; main M1 vs M3-QLoRA ranking is an empirical comparison, not a derivation that reduces to its inputs.

  1. self definitional [Abstract; §IV-B; §V-I; §VI-C; §VII-B]
    "Blockchain telemetry shows that lifecycle gas, cost, latency, and throughput differences are driven by submitted off-chain decision profiles rather than changes in fraud logic. ... the blockchain layer only recorded the already produced off-chain decision, applied the corresponding request-lifecycle operations, and stored execution telemetry for post-hoc audit. ... The blockchain component records already-formed off-chain decisions rather than performing fraud detection"

    The architecture defines the blockchain substrate as a passive recorder of already-formed off-chain APPROVE/BLOCK actions (no fraud scoring, no decision revision). Under that definition, differences in lifecycle gas, cost, latency, and throughput across M1–M3 must equal differences in the submitted decision profiles (review and finalization counts). Reporting this as an empirical result is tautological: the conclusion is identical to the design premise.

full rationale

This is a controlled synthetic systems comparison, not a first-principles derivation. Thresholds and coarse-state boundaries are selected on validation under an explicit legitimate-FPR cap (R_0.10,1.00) and frozen before deployment replay—standard ML operating-point selection, not a fitted parameter renamed as an independent prediction of the same quantity. The main claim (QLoRA becomes much more usable than zero-shot but mainly approaches rather than outperforms lower-cost M1) is supported by separate validation and labeled-deployment metrics (Table VII) on a drifted 100k corpus; those numbers are not forced by construction from the fit. The only clear circular step is the blockchain telemetry claim: the layer is defined to record already-formed off-chain decisions without revising fraud logic, so gas/latency/throughput differences must reflect submitted decision profiles by design. The paper states this design openly and treats the telemetry as confirmation rather than a surprising prediction; it is not load-bearing for the M1–M3 ranking. Self-citation of the authors’ prior conference paper [15] is present as lineage but is not used as a uniqueness theorem or to forbid alternatives. No uniqueness-from-authors, ansatz-via-citation, or renaming-known-result patterns. Overall circularity is minor and non-central.

Assumptions & free parameters 7 free parameters · 5 assumptions · 3 invented entities

The central ranking claim rests on synthetic data design choices, a fixed operating rule, configuration-specific score boundaries fitted on validation, and the assumption that off-chain risk signals can be compared fairly under one frozen policy and local chain. No new physical entities; invented constructs are architectural (five-state policy, two-zone refinement, AVG6, synthetic corpora). Free parameters are the usual ML/systems knobs that determine reported FPR/recall and gas profiles.

free parameters (7)
  • R0.10,1.00 operating rule (legitimate FPR ceiling 0.10, soft-fraud recall target 1.00)
    Validation-time threshold selection rule that freezes all downstream policy boundaries; directly shapes reported balance metrics.
  • Coarse-state boundaries b1/b2/b3 and low/high refinement thresholds per config
    Fitted/selected on validation evidence (Table VI); map continuous scores into NO/MAYBE/YES and control block rates.
  • Synthetic OOB hard-gate bounds (latency>110ms, bandwidth>100Mbps, power>20dBm, signal outside [-105,-25] dBm, duration>6
    Design-specific operating envelope that defines HARD_FRAUD and removes hard cases from statistical fitting.
  • AVG6 ensemble membership and arithmetic averaging
    Chosen after validation screening among many candidates; defines the M1 risk signal and M2 base streams.
  • Federated meta-learning hyperparameters (7 device-type clients, 12 rounds, 1 local epoch, SGD lr 0.01)
    Hand-set federation schedule that produces the M2 meta-probability.
  • M3-QLoRA training/inference settings (Qwen2.5-7B, 4-bit NF4, max length 768, LEGITIMATE/FRAUD labels)
    Adaptation choices that determine the softmax fraud probability used as M3-QLoRA risk signal.
  • Training/deployment fraud prevalence and soft/hard mix (train 15% fraud 60/40 soft/hard; deploy 5% fraud 10/90 soft/hard
    Generator settings that create the controlled drift-replay gap the main FPR results depend on.
assumptions (5)
  • domain assumption Synthetic profile-based trajectories with an eight-type soft/hard fraud taxonomy adequately support controlled comparison of deployment-stage risk sources.
    Stated throughout §V-A and limitations; evaluation is only as strong as this proxy.
  • domain assumption Deterministic OOB hard-fraud should be separated from inferential soft-fraud scoring, with hard cases excluded from supervised fitting.
    Core design in §III and §V-E; drives perfect hard recall and focuses residual error on soft fraud.
  • ad hoc to paper Heterogeneous risk sources can be compared fairly when request substrate, policy resolver, and blockchain audit path are held fixed.
    Methodological premise of the M1–M3 design (§IV-A); enables attribution of outcome differences to risk signals.
  • domain assumption Local Ganache Ethereum-compatible execution with consensus-inspired validator-assignment modes is sufficient to study lifecycle gas/latency/throughput drivers.
    §V-I–J and §VII-B; authors note these are not native production consensus measurements.
  • standard math Standard supervised learning, calibration, federated averaging-style aggregation, and QLoRA sequence classification behave as expected under the described protocols.
    Background ML/FL/LLM tooling assumptions used without re-derivation.
invented entities (3)
  • Five-state policy with two-zone refinement (HARD_FRAUD, NO, MAYBE_LOW, MAYBE_HIGH, YES)
    purpose: Map continuous risk scores into operational APPROVE/BLOCK actions with review-mediated ambiguous zones.
    Architectural construct of this framework; not an external physical entity, but load-bearing for all reported decision profiles.
  • AVG6 hybrid ensemble (LightGBM, XGBoost, GBM, RF, MLP, FT-Transformer arithmetic average)
    purpose: Centralized non-hard risk backbone for M1 and base probability streams for M2.
    Paper-specific retained backbone after candidate search; independent only insofar as member models are standard.
  • Controlled synthetic telecom/IoT training and 100k deployment-replay corpora
    purpose: Enable leakage-aware training, hard/soft separation, and post-training drift replay without real operator data.
    Generated for this study; all quantitative claims rest on these corpora.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Blockchain-Linked Auditable Decision Management for Telecom/IoT Fraud-Control Requests." pith.science (2026). https://pith.science/paper/IYUNKMN6

@misc{pith2026260709259,
  author       = {Pith},
  title        = {Pith review of: Blockchain-Linked Auditable Decision Management for Telecom/IoT Fraud-Control Requests},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IYUNKMN6}},
  note         = {Machine review of arXiv:2607.09259}
}
read the original abstract

Telecom fraud-control studies often stop at detector-level classification, but deployment use requires request-level policy resolution, lifecycle traceability, and auditability. This paper reframes fraud control as blockchain-linked auditable decision management for synthetic telecom/IoT fraud-control requests, and its main result is that the QLoRA-tuned LLM branch becomes much more usable than zero-shot prompting but mainly approaches, rather than outperforms, a lower-cost centralized ensemble. The framework maps each synthetic deployment record to a managed request, blocks explicit out-of-boundary cases through a deterministic hard-fraud gate, scores non-hard requests using centralized ML (M1), federated meta-learning (M2), or LLM-family risk sources (M3), and resolves actions through a shared five-state policy, two-zone refinement mechanism, and local Ethereum-compatible audit layer. Evaluation uses separate synthetic training data and a 100,000-record deployment replay corpus, so the study should be read as controlled drift-replay evidence rather than field validation or proof of live deployability. On validation, M1 gives the strongest balance, with legitimate-request FPR 0.0890 under the 0.10 operating cap and soft-fraud recall 0.8341. On labeled deployment replay, however, the legitimate-FPR gap becomes large: M1 rises to 0.1646 and M3-QLoRA to 0.1801, while M3-QLoRA reduces the M3-Base legitimate FPR from 0.3915 and reaches 0.8240 soft-fraud recall. Blockchain telemetry shows that lifecycle gas, cost, latency, and throughput differences are driven by submitted off-chain decision profiles rather than changes in fraud logic.

Figures

Figures reproduced from arXiv: 2607.09259 by the authors.

Figure 1
Figure 1. End-to-end workflow of the proposed blockchain-linked auditable decision-management framework for telecom/IoT fraud-control requests. The workflow separates training-side artifact development, configuration-specific risk-source generation, shared five-state policy resolution, and blockchain-linked execution/audit recording. The final off-chain action ai,k is forwarded to the shared blockchain-linked execution and au… view at source ↗
Figure 2
Figure 2. Deployment decision-state and outcome profiles across off￾chain risk configurations. (a) Percentage and count of deployment requests assigned to each decision state under M1, M2, M3-Base, and M3-QLoRA. (b) Corresponding final operational outcomes. Percent￾ages are computed over the full 100,000-record deployment replay. signal. M3-Base retained more discrete and higher zero-shot probability boundaries, which is cons… view at source ↗
Figure 3
Figure 3. Pairwise blocked-request-set overlap across deployment con￾figurations. Each bar reports the Jaccard overlap, 100 × |Bi ∩ Bj |/|Bi ∪ Bj |, between the requests blocked by two configurations. The percentages measure similarity between blocked-request sets, not the percentage of all deployment requests that were blocked. of non-hard cases into MAYBE_HIGH, indicating that zero￾shot structured LLM probabilities are less… view at source ↗
Figures from the paper (4 more)
Figure 2
Figure 2. Figure 2: M3-QLoRA substantially improves over M3-Base by [PITH_FULL_IMAGE:figures/full_fig_p012_2.png]
Figure 4
Figure 4. Figure 4: Component-level soft-fraud miss profile on the labeled deployment re [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 2
Figure 2. Figure 2: As shown in Fig. 5, the number of review transactions [PITH_FULL_IMAGE:figures/full_fig_p013_2.png]
Figure 5
Figure 5. Figure 5: Blockchain-linked lifecycle workload per validator-assignment mode. The bars report review-resolution, approved-request final￾ization, and blocked-request retention counts for each deployment configuration. largest review workload because many non-hard requests are rou…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

28 extracted references · 1 linked inside Pith

  1. [1]

    Exploiting spatial-temporal behavior patterns for fraud detection in telecom networks,

    G. Chu, J. Wang, Q. Qi, H. Sun, S. Tao, H. Yang, J. Liao, and Z. Han, “Exploiting spatial-temporal behavior patterns for fraud detection in telecom networks,”IEEE Transactions on Dependable and Secure Computing, vol. 20, no. 6, pp. 4564–4577, Nov./Dec. 2023

    2023

  2. [2]

    Telecommunications fraud increased 12% in 2023 equating to an estimated $38.95 bil- lion lost to fraud,

    Communications Fraud Control Association, “Telecommunications fraud increased 12% in 2023 equating to an estimated $38.95 bil- lion lost to fraud,” https://cfca.org/telecommunications-fraud-increased- 12-in-2023-equating-to-an-estimated-38-95-billion-lost-to-fraud/, Nov. 2023, accessed: 15 May 2026

    2023

  3. [3]

    The telecom fraud landscape in 2026: How the industry is fighting back,

    Transaction Network Services, “The telecom fraud landscape in 2026: How the industry is fighting back,” https://tnsi.com/resource/com/the- telecom-fraud-landscape-in-2026-how-the-industry-is-fighting-back- blog/, Jan. 2026, accessed: 8 June 2026

    2026

  4. [4]

    Wangiri fraud detection: A comprehensive approach to unlabeled telecom data,

    A. Balouchi, M. Abdollahi, A. Eskandarian, K. K. Pour Kerman, E. Majd, N. Azouji, and A. Baniasadi, “Wangiri fraud detection: A comprehensive approach to unlabeled telecom data,”Future Internet, vol. 18, no. 1, p. 15, 2026

    2026

  5. [5]

    On the road to 6G: Visions, requirements, key technologies, and testbeds,

    C.-X. Wang, X. You, X. Gao, X. Zhu, Z. Li, C. Zhang, H. Wang, Y . Huang, Y . Chen, H. Haas, J. S. Thompson, E. G. Larsson, M. Di Renzo, W. Tong, P. Zhu, X. Shen, H. V . Poor, and L. Hanzo, “On the road to 6G: Visions, requirements, key technologies, and testbeds,” IEEE Communications Surveys & Tutorials, vol. 25, no. 2, pp. 905–974, 2023

    2023

  6. [6]

    A comprehensive survey on resource management in 6G network based on Internet of Things,

    S. S. Sefati, A. Ul Haq, Nidhi, R. Craciunescu, S. Halunga, A. Mihovska, and O. Fratu, “A comprehensive survey on resource management in 6G network based on Internet of Things,”IEEE Access, vol. 12, pp. 113 741–113 784, 2024

    2024

  7. [7]

    Artificial intelligence advances in anomaly detection for telecom networks,

    E. Edozie, A. N. Shuaibu, B. O. Sadiq, and U. K. John, “Artificial intelligence advances in anomaly detection for telecom networks,” Artificial Intelligence Review, vol. 58, no. 4, p. 100, 2025

    2025

  8. [8]

    Current time series anomaly detection benchmarks are flawed and are creating the illusion of progress,

    R. Wu and E. J. Keogh, “Current time series anomaly detection benchmarks are flawed and are creating the illusion of progress,”IEEE Transactions on Knowledge and Data Engineering, vol. 35, no. 3, pp. 2421–2429, Mar. 2023

    2023

Show all 28 references
  1. [9]

    Leakage-aware benchmarking of temporal data prepa- ration for hourly energy forecasting,

    M. M. Sesay, “Leakage-aware benchmarking of temporal data prepa- ration for hourly energy forecasting,”Energy Conversion and Manage- ment: X, vol. 30, p. 101791, 2026

    2026

  2. [10]

    European space agency benchmark for anomaly detection in satellite telemetry,

    K. Kotowski, C. Haskamp, J. Andrzejewski, B. Ruszczak, J. Nalepa, D. Lakey, P. Collins, A. Kolmas, M. Bartesaghi, J. Martinez-Heras, and G. De Canio, “European space agency benchmark for anomaly detection in satellite telemetry,” arXiv preprint arXiv:2406.17826, 2024, revised 2025

    2024 arXiv

  3. [11]

    Beyond the individual: An improved telecom fraud detection approach based on latent synergy graph learning,

    J. Wu, R. Hu, D. Li, L. Ren, Z. Huang, and Y . Zang, “Beyond the individual: An improved telecom fraud detection approach based on latent synergy graph learning,”Neural Networks, vol. 169, pp. 20–31, Jan. 2024

    2024

  4. [12]

    Federated learning for telecom fraud detection: A privacy-preserving approach to overcoming data fragmentation and enhancing security,

    K. Hasan, M. N. Hosen, and K. Saha, “Federated learning for telecom fraud detection: A privacy-preserving approach to overcoming data fragmentation and enhancing security,”European Journal of Theoretical and Applied Sciences, vol. 2, no. 6, pp. 99–109, Nov. 2024

    2024

  5. [13]

    Telecom fraud detection based on large language models: A multi-role, multi-layer prompting strategy,

    J. Ding and H. Zhou, “Telecom fraud detection based on large language models: A multi-role, multi-layer prompting strategy,”Applied Sciences, vol. 16, no. 1, p. 544, Jan. 2026

    2026

  6. [14]

    Blockchain in communication networks: A comprehensive review,

    Q. Mamun, Z. Pan, and J. Wu, “Blockchain in communication networks: A comprehensive review,”IET Blockchain, vol. 6, no. 1, p. e70031, Jan. 2026

    2026

  7. [15]

    Hybrid telecom fraud detection with machine learning, large language models, and blockchain,

    S. Changizi, N. Mohammadzadeh, and M. Shojafar, “Hybrid telecom fraud detection with machine learning, large language models, and blockchain,” inProceedings of the 2025 IEEE International Conference on High Performance Computing and Communications (HPCC), 2025, pp. 1094–1102

    2025

  8. [16]

    Telecom fraud detection via Hawkes-enhanced sequence model,

    Y . Jiang, G. Liu, J. Wu, and H. Lin, “Telecom fraud detection via Hawkes-enhanced sequence model,”IEEE Transactions on Knowledge and Data Engineering, 2022, early access

    2022

  9. [17]

    GAT-COBO: Cost-sensitive graph neural network for telecom fraud detection,

    X. Hu, H. Chen, J. Zhang, H. Chen, S. Liu, X. Li, Y . Wang, and X. Xue, “GAT-COBO: Cost-sensitive graph neural network for telecom fraud detection,”IEEE Transactions on Big Data, vol. 10, no. 4, pp. 528–542, Aug. 2024

    2024

  10. [18]

    Ai anomaly detection for cloudified mobile core architectures,

    F. Michelinakis, J. S. Pujol-Roig, S. Malacarne, M. Xie, T. Dreibholz, S. Majumdar, W. Y . Poe, G. Patounas, C. Guerrero, A. Elmokashfi, and V . Theodorou, “Ai anomaly detection for cloudified mobile core architectures,”IEEE Transactions on Network and Service Management, vol....

    1976

  11. [19]

    Real-time adaptive anomaly detection in industrial iot envi- ronments,

    M. Raeiszadeh, A. Ebrahimzadeh, R. H. Glitho, J. Eker, and R. A. F. Mini, “Real-time adaptive anomaly detection in industrial iot envi- ronments,”IEEE Transactions on Network and Service Management, vol. 21, no. 6, pp. 6839–6856, Dec. 2024

    2024

  12. [20]

    Federated learning for 6G security: A survey on threats, solutions, and research directions,

    C. de Alwis, O. Aouedi, J. Xu, S. Wang, Y . Siriwardhana, T. Hewa, E. Zeydan, C. Sandeepa, and M. Liyanage, “Federated learning for 6G security: A survey on threats, solutions, and research directions,” IEEE Communications Surveys & Tutorials, vol. 28, pp. 4883–4914, Feb. 2026

    2026

  13. [21]

    Toward securing federated learning against poisoning attacks in zero touch b5g networks,

    S. Ben Saad, B. Brik, and A. Ksentini, “Toward securing federated learning against poisoning attacks in zero touch b5g networks,”IEEE Transactions on Network and Service Management, vol. 20, no. 2, pp. 1612–1624, Jun. 2023

    2023

  14. [22]

    Telecom fraud recognition based on large language model neuron selection,

    L. Jiang, C. Zhang, X. Qin, Y . Zhou, G. Huang, H. Li, and J. Li, “Telecom fraud recognition based on large language model neuron selection,”Mathematics, vol. 13, no. 11, p. 1784, May 2025

    2025

  15. [23]

    Nx- GenT: A decentralized trust system for B5G and 6G networks based on blockchain and smart contracts,

    C. Núñez-Gómez, V . Garcia-Font, H. Rifà-Pous, and M. Asad, “Nx- GenT: A decentralized trust system for B5G and 6G networks based on blockchain and smart contracts,”Blockchain: Research and Applications, p. 100447, 2026

    2026

  16. [24]

    Blockchain for dynamic spectrum access and network slicing: A review,

    S. T. Muntaha, P. I. Lazaridis, M. Hafeez, Q. Z. Ahmed, F. A. Khan, and Z. D. Zaharis, “Blockchain for dynamic spectrum access and network slicing: A review,”IEEE Access, vol. 11, pp. 17 922–17 944, 2023

    2023

  17. [25]

    Distributed-proof-of-sense: Blockchain consensus mechanisms for detecting spectrum access vio- lations of the radio spectrum,

    P. Fernando, K. Dadallage, T. Gamage, C. Seneviratne, A. Braeken, A. Madanayake, and M. Liyanage, “Distributed-proof-of-sense: Blockchain consensus mechanisms for detecting spectrum access vio- lations of the radio spectrum,”IEEE Transactions on Cognitive Com- munications and ...

    2023

  18. [26]

    Trust aware and explainable access control for Internet of Medical Things: A lightweight hybrid blockchain approach,

    M. Alashwal, U. Waheed, M. Algarni, H. Jamshed, M. Zwawi, Y . Man- soor, and A. Hussain, “Trust aware and explainable access control for Internet of Medical Things: A lightweight hybrid blockchain approach,” IEEE Internet of Things Journal, 2026, early access

    2026

  19. [27]

    TRADE-5G: A blockchain- based transparent and secure resource exchange for 5G network slicing,

    E.-h. Diallo, K. Al Agha, and S. Martin, “TRADE-5G: A blockchain- based transparent and secure resource exchange for 5G network slicing,” Blockchain: Research and Applications, vol. 6, p. 100246, 2025

    2025

  20. [28]

    The OPS-SAT benchmark for detecting anomalies in satellite telemetry,

    B. Ruszczak, K. Kotowski, D. Evans, and J. Nalepa, “The OPS-SAT benchmark for detecting anomalies in satellite telemetry,”Scientific Data, vol. 12, p. 710, 2025

    2025

Pith tools

Reviewed July 13, 2026 · model on record in the stance chip above.