Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

Deciphering the global production network from cross-border firm transactions

T0 review · 4 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read A directed product network built from one billion cross-border firm transactions shows that upstream linkages—who supplies whom—best predict which products countries begin to export.

desk verdict The submission packet is broken: the abstract describes a major empirical production-network study, but the attached full text is an unrelated wireless-jamming paper, so none of the claims can be verified. read the letter →

arxiv 2508.12315 v4 pith:OGMPEZPC submitted 2025-08-17 econ.GN q-fin.EC

classification econ.GNq-fin.EC
keywords globalproductionnetworksupplychainsfirm-leveltransactionsproductdiversificationinput-outputlinkagessciencebackwardexportpatterns
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

This paper tries to map the global production network at product level without relying on national input-output tables, by converting a billion cross-border transactions between 20 million firms into a directed network covering 1,200 products. If the construction is sound, it would provide the first empirical, transaction-derived view of who supplies what to whom, and would let researchers test how network position drives industrial diversification. The paper's central empirical claim is that a country's or product's position in this network predicts diversification into new products, with backward (upstream) linkages—connections to suppliers—carrying stronger predictive weight than forward linkages to customers. It also reports that product communities line up with textiles, chemicals and food, and machinery and metals, and that the network structurally agrees with an LLM-generated reference network and with manually mapped battery and semiconductor supply chains.

What carries the argument

The central object is the directed product-level network built by aggregating 1 billion firm-to-firm cross-border transactions and assigning them to products. This network carries the argument: the three product communities, the hub positions of European industrial nations and China, and the forward and backward linkage measures are all computed from it, and the diversification regressions then test whether network position predicts entry into new product lines.

What would settle it

Take the same transaction dataset, remap firms to products using a different concordance or raise the minimum transaction threshold, and check whether the same three product clusters appear and whether backward linkages still out-predict forward linkages in the diversification regressions; if the cluster structure or the sign and strength of the linkage coefficients change materially, the network construction step is driving the results rather than the real supply-chain structure.

Watch

Extended reading notes

Core claim

The paper claims it can construct a directed global input network at product level—nodes are roughly 1,200 products, edges are inferred from cross-border transactions between 20 million firms—and that this network shows three large communities, with European industrial countries and China as hubs for critical intermediate products. It then claims that both forward and backward linkage centrality predict whether a country starts exporting new products, with backward (upstream) linkages the stronger predictor. The paper also reports structural agreement with an independent LLM-generated network (AIPNET) and with manually mapped electric vehicle battery and semiconductor supply chains, which it offers as evidence that the transaction-derived network captures real production relationships.

Load-bearing premise

The load-bearing premise is that firm-to-firm transaction records can be faithfully converted into a product-level input network—a step the abstract does not describe, and which requires deciding how transactions map to products, how prices reflect real inputs, and which transactions to keep.

Editorial extensions

If this is right

  • If the network holds, policy analysts can identify critical intermediate products and their concentrated producers from transaction data rather than from survey-based tables.
  • The backward-linkage result implies that access to upstream suppliers, more than customer proximity, is what enables a country to enter new export lines.
  • The three product clusters offer a data-driven taxonomy of global production that could be compared across years to see how supply chains reorganize.
  • Agreement with AIPNET and manual maps suggests LLM-based and transaction-based supply chain mapping can cross-validate each other.

Reading between the lines

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

  • A natural next test is whether the backward-linkage advantage survives when only arms-length transactions are used, since transfer prices within multinational groups could inflate certain upstream links.
  • The same network could be used to simulate shock propagation, for example which products would be hit if a hub country stopped exporting, a question the paper does not itself run.
  • Because the abstract reports agreement with AIPNET but not its strength, a quantified agreement statistic would tell whether the two methods converge on the same structure or merely share broad clusters.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper claims to infer a directed product-level input network from international firm-level transaction data covering 20 million global firms and 1 billion cross-border transactions, yielding key input relationships for over 1,200 products. The abstract further reports three large product clusters (textiles, chemicals and food, machinery and metals), country-level dominance patterns in intermediate products, and a comparative test showing that both forward and backward linkages predict country-product diversification, with stronger overall evidence for backward (upstream) linkages. The abstract also reports structural similarity between the inferred network and AIPNET, an LLM-generated reference network, and strong agreement with manually mapped electric vehicle battery and semiconductor supply chains. However, the full text supplied with the submission is arXiv:2508.12320, a paper on wireless jamming identification, and contains none of the production-network methods, data description, results, or validation. As a result, the central claims cannot currently be checked against any detailed evidence.

Significance. If the results hold, the paper would provide the first global product-level input network inferred from firm-level cross-border transaction data at this scale, and a direct empirical test of forward versus backward linkage effects on diversification. Such a contribution would be valuable for trade and development economics, and the use of actual transaction records rather than aggregate input-output tables is a genuine strength of the proposed approach. The claims are also falsifiable in principle: the three-cluster structure, the dominance rankings, and the linkage-diversification regressions could all be tested by other researchers if the data and code were shared. At present, however, none of this evidence is available in the submitted manuscript, so the significance cannot be assessed beyond the level of the abstract.

major comments (4)
  1. [Full text (supplied document)] The full text provided with this submission is arXiv:2508.12320, titled "Jamming Identification with Differential Transformer for Low-Altitude Wireless Networks," and it contains no content related to production networks, firm transactions, product clustering, or backward/forward linkages. Because the actual methods and results are absent, every quantitative claim in the abstract is currently unverifiable. This is a load-bearing issue: the transformation from firm-level transactions to a directed product network, the clustering procedure, and the linkage-diversification analysis are all unavailable for inspection.
  2. [Abstract, second sentence] The abstract states only that the data are "transforming this data to a directed network" with no description of how firm-level transactions are aggregated, cleaned, mapped to products, or filtered. The paper needs to specify the product-attribution rule, whether intra-firm or transfer-priced transactions are included, what firm or country thresholds are applied, and how edge weights are normalized before computing forward and backward linkage statistics. Without this information, the reported three clusters and the stronger backward-linkage result could be artifacts of classification or filtering choices rather than economic structure.
  3. [Abstract, final sentence (AIPNET comparison)] AIPNET is introduced as a reference network generated via LLM queries, but the abstract does not state who generated AIPNET, what prompts or data were used, or how "structural similarities" were measured. If AIPNET was constructed by the same authors, the comparison is not a fully independent external validation, and the abstract should say so explicitly. The claim of structural similarity needs a quantitative metric, a null model, and a statement of whether the comparison is more than visual or qualitative.
  4. [Abstract, linkage-diversification finding] The claim that "Both forward and backward linkages are predictive of country-product diversification patterns, with stronger overall evidence for backward (upstream) linkages" is a comparative empirical statement, but the abstract reports no regression specification, no coefficient magnitudes, no standard errors, and no robustness checks. Even allowing for the brevity of an abstract, the full text must provide these details; currently the full text supplies none of them, so the central empirical conclusion has no evidentiary support in the submission.
minor comments (4)
  1. [Abstract, first sentence] "20m global firms" should be written "20 million global firms" to avoid ambiguity with meters or other units.
  2. [Abstract, first sentence] The time span covered by "1 billion cross-border transactions" is not stated; the paper should specify the observation window and whether the transactions are unique firm-to-firm relationships or individual shipment records.
  3. [Abstract, final sentence] The acronym AIPNET is not defined; the abstract should expand it or at least describe what the letters stand for, and a reference to the LLM-based construction method should be provided.
  4. [Abstract, final sentence] The manually mapped electric vehicle battery and semiconductor supply chains are mentioned without references or a description of how the mapping was performed; the manuscript should cite the underlying sources and define the comparison procedure.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity can be demonstrated from the available abstract; the supplied full text is an unrelated paper, preventing inspection of the derivation chain.

full rationale

No circular step can be established from the provided evidence. The abstract claims an empirical inference from firm-level transaction data to a directed product network, and then tests whether forward and backward linkages predict country-product diversification. Nothing in the abstract defines the linkage measures in terms of the diversification outcome, nor does it state that the network was fitted to the outcome it later predicts. The AIPNET comparison is presented as a separate reference network generated via LLM queries; it is not claimed to be an input to the network construction, so finding structural similarity with it is a convergent-validity check rather than a derivation from it. The manually mapped electric vehicle battery and semiconductor supply chains are likewise external reference points, not fitted parameters. The most significant issue is that the full text supplied in the submission package is a completely different paper on wireless jamming identification, so the actual derivation chain, including the firm-to-product aggregation, cannot be inspected at all. However, lack of verifiability is not circularity: the rules require quoting a specific reduction where a prediction is equivalent to its inputs by construction, and no such reduction is available here. Accordingly, the circularity score is 0.

Assumptions & free parameters 1 free parameters · 3 assumptions · 1 invented entities

Abstract-only review. No free parameters can be identified from the abstract with certainty; the entries above are the most likely load-bearing choices. The axioms are domain assumptions stated or implied in the abstract. AIPNET is an introduced benchmark entity without independent evidence in the abstract.

free parameters (1)
  • Network inference thresholds and aggregation rules = Not stated in abstract
    The abstract does not specify how firm-level transactions are converted into product-level input links; any thresholds, filters, or normalization constants would be free parameters.
assumptions (3)
  • domain assumption Firm-level transaction data are representative of global production and product-level input flows
    The abstract assumes the transaction sample can stand in for the full input-output structure of the world economy.
  • domain assumption Product-level input requirements can be inferred from observed cross-border transactions
    The transformation 'transforming this data to a directed network' presumes that transaction flows map cleanly to input coefficients.
  • domain assumption AIPNET, generated via LLM queries, is a structurally meaningful reference network
    The abstract uses AIPNET as a comparison benchmark without stating how it was generated or validated.
invented entities (1)
  • AIPNET
    purpose: A reference network generated via LLM queries, used to validate the structure of the inferred production network
    The abstract presents AIPNET as a benchmark but gives no independent falsifiable evidence for its validity; it is introduced within this study.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Deciphering the global production network from cross-border firm transactions." pith.science (2026). https://pith.science/paper/OGMPEZPC

@misc{pith2026250812315,
  author       = {Pith},
  title        = {Pith review of: Deciphering the global production network from cross-border firm transactions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OGMPEZPC}},
  note         = {Machine review of arXiv:2508.12315}
}
read the original abstract

Critical for policy-making and business operations, the study of global supply chains has been severely hampered by a lack of detailed data. Here we harness international firm-level transaction data covering 20m global firms, and 1 billion cross-border transactions, to infer key inputs for over 1200 products. Transforming this data to a directed network, we find that products are clustered into three large groups including textiles, chemicals and food, and machinery and metals. European industrial nations and China dominate critical intermediate products such as metals, common components and tools, while industrial complexity is highly correlated with embeddedness in densely connected supply chains. Both forward and backward linkages are predictive of country-product diversification patterns, with stronger overall evidence for backward (upstream) linkages. Finally, we find structural similarities with AIPNET, a reference network generated via LLM queries, and strong linkages between products identified in manually-mapped electric vehicle battery and semiconductor supply chains.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Synthetic supply networks

    econ.GN 2026-08 conditional novelty 6.0 of 10

    A new pipeline generates synthetic firm-level supply networks that match selected empirical network statistics and aggregate exactly to national input-output tables using only public data.

Reference graph

Works this paper leans on

47 extracted references · 44 canonical work pages · cited by 1 Pith paper

  1. [1]

    Joint UA V de- ployment and space-time-frequency resource allocation for low-altitude economy,

    Y . Li, W. Wang, C. Zhang, Y . Huang and D. Niyato, "Joint UA V de- ployment and space-time-frequency resource allocation for low-altitude economy,"IEEE Wireless Commun. Lett., early access, 2025, DOI: 10.1109/LWC.2025.3579883

  2. [2]

    Ubiquitous UA V com- munication enabled low-altitude economy: applications, techniques, and 3GPP’s efforts,

    D. He, W. Yuan, J. Wu and R. Liu, "Ubiquitous UA V com- munication enabled low-altitude economy: applications, techniques, and 3GPP’s efforts,"IEEE Netw., early access, 2025, DOI: 10.1109/MNET.2025.3574922

  3. [3]

    Co-design of sensing, communications, and control for low-altitude wireless net- works,

    H. Jin, J. Wu, W. Yuan, F. Liu, and Y . Cui, “Co-design of sensing, communications, and control for low-altitude wireless net- works,”IEEE Trans. Mobile Comput., early access, 2025, DOI: 10.1109/TMC.2025.3581616

  4. [4]

    Deceiving reactive jamming in dynamic wireless sensor networks: a deep reinforcement learning based approach,

    C. Zhang, T. Mao, Z. Xiao, R. Liu and X. -G. Xia, “Deceiving reactive jamming in dynamic wireless sensor networks: a deep reinforcement learning based approach,” inProc. GLOBECOM 2023, (Kuala Lumpur, Malaysia), 2023, pp. 4455–4460

  5. [5]

    Jamming detection in broadband frequency hopping systems based on multi-segment signals spectrum clustering,

    H. Xu, Y . Cheng and P. Wang, “Jamming detection in broadband frequency hopping systems based on multi-segment signals spectrum clustering,”IEEE Access, vol. 9, pp. 29980–29992, 2021

  6. [6]

    Time-frequency component-aware convolutional neural network for wireless interference classification,

    P. Wang, Y . Cheng, G. Shang, J. Wang and S. Li, “Time-frequency component-aware convolutional neural network for wireless interference classification,”IEEE Wireless Commun. Lett., vol. 11, no. 12, pp. 2487– 2491, Dec. 2022

  7. [7]

    UA V anti-jamming communications with power and mobility control,

    H. Wang, G. Ding, J. Chen, Y . Zou and F. Gao, “UA V anti-jamming communications with power and mobility control,”IEEE Trans. Wire- less Commun., vol. 22, no. 7, pp. 4729–4744, July 2023

  8. [8]

    Physical-layer security enhancement for SIMO- MBM systems,

    T. Mao and Z. Wang, “Physical-layer security enhancement for SIMO- MBM systems,” inProc. GLOBECOM 2018, (Abu Dhabi, United Arab Emirates), 2018, pp. 1–6

Show all 47 references
  1. [9]

    Reinforcement learning-based dynamic anti-jamming power control in UA V networks: An effective jamming signal strength based approach,

    N. Ma et al., “Reinforcement learning-based dynamic anti-jamming power control in UA V networks: An effective jamming signal strength based approach,”IEEE Commun. Lett., vol. 26, no. 10, pp. 2355–2359, Oct. 2022

  2. [10]

    Bring globality into convolutional neural networks for wireless interference classification,

    P. Wang, Y . Cheng, B. Dong and Q. Peng, “Bring globality into convolutional neural networks for wireless interference classification,” IEEE Wireless Commun. Lett., vol. 11, no. 3, pp. 538–542, Mar. 2022

  3. [11]

    Data-driven deep learning for automatic modulation recognition in cognitive radios,

    Y . Wang, M. Liu, J. Yang and G. Gui, “Data-driven deep learning for automatic modulation recognition in cognitive radios,”IEEE Trans. V eh. Technol., vol. 68, no. 4, pp. 4074–4077, Apr. 2019

  4. [12]

    Learning dynamic computing resource allocation in convolutional neural networks for wireless interference identification,

    P. Wang, Y . Cheng, Q. Peng, J. Wang and S. Li, “Learning dynamic computing resource allocation in convolutional neural networks for wireless interference identification,”IEEE Trans. V eh. Technol., vol. 72, no. 7, pp. 8770–8782, Jul. 2023

  5. [13]

    Range dimensional monopulse approach with FDA-MIMO radar for mainlobe deceptive jamming suppression,

    M. Tan, J. Gong and C. Wang, “Range dimensional monopulse approach with FDA-MIMO radar for mainlobe deceptive jamming suppression,” IEEE Antennas Wireless Propag. Lett., vol. 23, no. 2, pp. 643–647, Feb. 2024

  6. [14]

    Signal fusion-based algo- rithms to discriminate between radar targets and deception jamming in distributed multiple-radar architectures,

    S. Zhao, L. Zhang, Y . Zhou and N. Liu, “Signal fusion-based algo- rithms to discriminate between radar targets and deception jamming in distributed multiple-radar architectures,”IEEE Sens. J., vol. 15, no. 11, pp. 6697–6706, Nov. 2015

  7. [15]

    Dis- crimination of active false targets based on hermitian distance for distributed multiple-radar architectures,

    Q. Li, L. Zhang, Y . Zhou, S. Zhao, N. Liu and J. Zhang, “Dis- crimination of active false targets based on hermitian distance for distributed multiple-radar architectures,”IEEE Access, vol. 7, pp. 71872– 71883, 2019

  8. [16]

    Radar detection and classification of jamming signals belonging to a cone class,

    M. Greco, F. Gini and A. Farina, “Radar detection and classification of jamming signals belonging to a cone class,”IEEE Trans. Signal Process., vol. 56, no. 5, pp. 1984–1993, May 2008

  9. [17]

    Discrimination between radar targets and deception jamming in distributed multiple- radar architectures,

    S. Zhao, Y . Zhou, L. Zhang, Y . Guo and S. Tang, “Discrimination between radar targets and deception jamming in distributed multiple- radar architectures,”IET Radar , Sonar Navigat., vol. 11, no. 7, pp. 1124– 1131, 2017

  10. [18]

    Research on anti- jamming technology of chaotic composite short range detection system based on underdetermined signal separation and spectral analysis,

    D. Wei, S. Zhang, S. Chen, H. Zhao, and L. Zhu, “Research on anti- jamming technology of chaotic composite short range detection system based on underdetermined signal separation and spectral analysis,”IEEE Access, vol. 7, pp. 42298–42308, 2019

  11. [19]

    Jamming efficiency analysis based on the range profile of target with chaff,

    Y . Zuo, L. Guo, W. Liu, and J. Ding, “Jamming efficiency analysis based on the range profile of target with chaff,”IEEE Access, vol. 9, pp. 13573–13589, 2021

  12. [20]

    Jamming recognition method based on the polarisation scattering characteristics of chaff clouds,

    Y . Liu,et al., “Jamming recognition method based on the polarisation scattering characteristics of chaff clouds,”IET Radar Sonar Navigation., vol. 11, no. 11, pp. 1689– 1699, 2017

  13. [21]

    A dynamic discretization approach for constructing decision trees with a continuous label,

    H. -W. Hu, Y . -L. Chen and K. Tang, “A dynamic discretization approach for constructing decision trees with a continuous label,”IEEE Trans. Knowl. Data Eng., vol. 21, no. 11, pp. 1505-1514, Nov. 2009

  14. [22]

    Twin support vector machines for pattern classification,

    Jayadeva, R. Khemchandani and S. Chandra, “Twin support vector machines for pattern classification,”IEEE Trans. Pattern Anal. Mach. Intell., vol. 29, no. 5, pp. 905-910, May 2007

  15. [23]

    Very deep convolutional networks for large-scale image recognition

    K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition”, inProc. ICLR 2015, pp. 1–14, 2015

  16. [24]

    Deep residual learning for image recognition,

    K. He,et al., “Deep residual learning for image recognition,” inProc. CVPR 2016, (Las Vegas, NV , USA), 2016, pp. 770–778

  17. [25]

    Rethinking the inception architecture for computer vision,

    C. Szegedy, V . Vanhoucke, S. Ioffe, J. Shlens and Z. Wojna, “Rethinking the inception architecture for computer vision,” inProc. CVPR 2016, (Las Vegas, NV , USA), 2016, pp. 2818–2826

  18. [26]

    Squeeze-and-excitation networks,

    J. Hu, L. Shen, S. Albanie, G. Sun and E. Wu, “Squeeze-and-excitation networks,” inProc. CVPR 2019, Sep. 2019, pp. 7132–7141

  19. [27]

    Learning-based wireless powered secure transmission,

    D. He, C. Liu, H. Wang and T. Q. S. Quek, “Learning-based wireless powered secure transmission,”IEEE Wireless Commun. Lett., vol. 8, no. 2, pp. 600–603, Apr. 2019

  20. [28]

    Deep learning-assisted demodulation for tera- Hertz communications under hybrid distortions,

    D. He and Z. Wang, “Deep learning-assisted demodulation for tera- Hertz communications under hybrid distortions,”IEEE Commun. Lett., vol. 26, no. 2, pp. 325–329, Feb. 2022

  21. [29]

    Transformer-empowered 6G intelligent networks: From massive MIMO processing to semantic communication,

    Y . Wang, Z. Gao, D. Zheng, S. Chen, D. Gunduz and H. V . Poor, “Transformer-empowered 6G intelligent networks: From massive MIMO processing to semantic communication,”IEEE Wireless Commun., vol. 30, no. 6, pp. 127-135, Dec. 2023

  22. [30]

    MmWave beam prediction under hand blockage,

    P. Wang, C. Sun, K. Ma, Y . Bai and Z. Wang, “MmWave beam prediction under hand blockage,”IEEE Wireless Commun. Lett., vol. 13, no. 12, pp. 3598–3602, Dec. 2024

  23. [31]

    Robust deep sensing through transfer learning in cognitive radio,

    Q. Peng, A. Gilman, N. Vasconcelos, P. C. Cosman and L. B. Milstein, “Robust deep sensing through transfer learning in cognitive radio,”IEEE Wireless Commun. Lett., vol. 9, no. 1, pp. 38–41, Jan. 2020

  24. [32]

    Adversarial attacks in modulation recognition with convolutional neural networks,

    Y . Lin, H. Zhao, X. Ma, Y . Tu and M. Wang, “Adversarial attacks in modulation recognition with convolutional neural networks,”IEEE Trans. Reliability, vol. 70, no. 1, pp. 389–401, Mar. 2021

  25. [33]

    Complex-valued networks for automatic modulation classification,

    Y . Tu, Y . Lin, C. Hou and S. Mao, “Complex-valued networks for automatic modulation classification,”IEEE Trans. V eh. Technol., vol. 69, no. 9, pp. 10085–10089, Sep. 2020

  26. [34]

    Automatic modulation classification based on decentralized learning and ensemble learning,

    X. Fu, G. Gui, Y . Wang, H. Gacanin and F. Adachi, “Automatic modulation classification based on decentralized learning and ensemble learning,”IEEE Trans. V eh. Technol., vol. 71, no. 7, pp. 7942–7946, Jul. 2022

  27. [35]

    Binary neural networks for wireless interference identification,

    P. Wang, Y . Cheng, B. Dong and G. Gui, “Binary neural networks for wireless interference identification,”IEEE Wireless Commun. Lett., vol. 11, no. 1, pp. 23–27, Jan. 2022

  28. [36]

    Convolutional neural network-based radar jamming signal classification with sufficient and limited samples,

    G. Shao, Y . Chen and Y . Wei, “Convolutional neural network-based radar jamming signal classification with sufficient and limited samples,”IEEE Access, vol. 8, pp. 80588–80598, 2020

  29. [37]

    JRNet: jamming recognition networks for radar compound suppression jamming signals,

    Q. Qu, S. Wei, S. Liu, J. Liang and J. Shi, “JRNet: jamming recognition networks for radar compound suppression jamming signals,”IEEE Trans. V eh. Technol., vol. 69, no. 12, pp. 15035–15045, Dec. 2020

  30. [38]

    Abandon locality: Frame-wise embedding aided transformer for automatic modulation recognition,

    Y . Chen, B. Dong, C. Liu, W. Xiong and S. Li, “Abandon locality: Frame-wise embedding aided transformer for automatic modulation recognition,”IEEE Commun. Lett., vol. 27, no. 1, pp. 327-331, Jan. 2023

  31. [39]

    Adversarial attack and defense on deep learning for air transportation communication jamming,

    M. Liu, Z. Zhang, Y . Chen, J. Ge and N. Zhao, “Adversarial attack and defense on deep learning for air transportation communication jamming,”IEEE Trans. Intell. Transp. Syst., vol. 25, no. 1, pp. 973-986, Jan. 2024

  32. [40]

    Adversarial attacks on deep-learning based radio signal classification,

    M. Sadeghi and E. G. Larsson, “Adversarial attacks on deep-learning based radio signal classification,”IEEE Wireless Commun. Lett., vol. 8, no. 1, pp. 213–216, Feb. 2019

  33. [41]

    Threat of adversarial attacks on DL-based IoT device identification,

    Z. Bao, Y . Lin, S. Zhang, Z. Li and S. Mao, “Threat of adversarial attacks on DL-based IoT device identification,”IEEE Internet Things J., vol. 9, no. 11, pp. 9012–9024, Jun. 2022

  34. [42]

    Explaining and harnessing adversarial examples,

    I. J. Goodfellow, J. Shlens, and C. Szegedy, “Explaining and harnessing adversarial examples,” inProc. ICLR 2015, (San Diego, USA), May 7–9, 2015, pp. 189–199

  35. [43]

    Distillation as a defense to adversarial perturbations against deep neural networks,

    N. Papernot, P. Mcdaniel, X. Wu, S. Jha, and A. Swami, “Distillation as a defense to adversarial perturbations against deep neural networks,” inProc. IEEE Symp. Security Privacy (SP), pp. 582–597, 2016

  36. [44]

    Feature denoising for improving adversarial robustness,

    C. Xie,et al., “Feature denoising for improving adversarial robustness,” inProc. CVPR 2019, (Long Beach, USA), pp. 501–509, 2019

  37. [45]

    Towards achieving adversarial robustness by enforcing feature consis- tency across bit planes,

    S. Addepalli, S. VivekB., A. Baburaj, G. Sriramanan, and R. V . Babu, “Towards achieving adversarial robustness by enforcing feature consis- tency across bit planes,” inProc. CVPR 2020, pp. 1017–1026, 2020

  38. [46]

    Differential transformer,

    T. Ye, L. Dong, Y . Xia, Y . Sun, Y . Zhu, G. Huang, F. Wei, “Differential transformer,” 2024,arXiv: 2410.05258

  39. [47]

    An image is worth 16x16 words: transformers for image recognition at scale,

    A. Dosovitskiy,et al., “An image is worth 16x16 words: transformers for image recognition at scale,” inProc. ICLR 2021, May 3–7, 2021, pp. 1–21

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.