Pith. sign in

REVIEW 3 major objections 6 minor 39 references

Detecting Zero-Day Attacks in Digital Substations via In-Context Learning

T0 review · 3 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read This paper claims that in-context learning lets a transformer detect zero-day attacks on IEC-61850 digital substation traffic with 85.8–100% accuracy at first encounter, without retraining.

desk verdict A real extension (distributional transformer) buried under an evaluation protocol that tunes on the test attacks; the zero-day numbers are likely optimistic but the approach deserves a serious referee. read the letter →

arxiv 2501.16453 v1 pith:DVFDOVCN submitted 2025-01-27 cs.LG cs.AI

Faizan Manzoor , Vanshaj Khattar , Akila Herath , Clifton Black , Matthew C Nielsen , Junho Hong , Chen-Ching Liu , Ming Jin This is my paper · ORCID
classification cs.LGcs.AI
keywords in-contextlearningIEC-61850intrusiondetectionsystemszero-dayattacksGPT-2transformerdistributionalmulti-mixing
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 argues that the in-context learning ability of transformers—the same mechanism that lets large language models adapt to new tasks from a few prompt examples—can be turned into an intrusion detection system for IEC-61850 digital substations. The authors train a GPT-2-style transformer on packets from five known attack classes plus normal traffic, augmented by synthetic classes, and at deployment feed it a window of recent packets labeled only by weak classifiers. The transformer predicts whether the newest packet is normal or an attack. On the ERENO dataset, the best variant (a distributional transformer consuming label probabilities rather than hard labels) achieves between 85.8% and 100% accuracy on four held-out zero-day attack types even before seeing any instance of them, and reaches 94.9–100% after a few in-context examples, while standard machine-learning baselines fail on at least one of these attacks. If correct, this gives substation operators a detector that can recognize novel attacks without retraining and with zero false positives on normal traffic.

What carries the argument

The framework has three moving parts. Multi-mixing generates synthetic attack classes by averaging randomly selected subsets of the five known attack-class feature sets, multiplying the class diversity available for training. Weak classifiers, trained on this augmented set, provide pseudo-labels for in-context packets; the distributional transformer (DTF) consumes the full probability distribution each weak classifier assigns across the synthetic classes, which the authors find gives better zero-shot accuracy and robustness than the hard-label variant. The transformer itself is a decoder-only GPT-2 model with causal attention, so its prediction for the query packet depends only on the preceding labeled packets in the window, which is exactly the temporal structure of a substation packet stream.

What would settle it

Run the same MDTF training on the five ID classes, then evaluate on attacks purposely engineered to lie outside the convex hull of the ID feature distributions (for example, adversarial perturbations of normal packets that cross the decision boundary while preserving protocol validity); if zero-shot accuracy falls below 80% while the weak classifiers remain above 50% accuracy, the multi-mixing manifold premise is falsified.

Watch

Extended reading notes

Core claim

The central discovery is that in-context learning transfers from language to network traffic: a transformer conditioned on a short history of packet-label pairs can flag an unseen attack class as anomalous, and can improve its detection of that attack as more of its packets appear in the context window, all without parameter updates. The mixed-trained distributional transformer (MDTF) was the strongest instantiation, with no failure cases across the four out-of-distribution attack types in Table 1 and perfect normal-traffic accuracy; it also respects the IEC-61850 3 ms message transmission time when run on a GPU with TensorRT. The paper frames this as evidence that transformer-based IDS can generalize to zero-day attacks where fixed-rule and fixed-classifier baselines do not.

Load-bearing premise

The claim rests on the premise that the synthetic attack classes produced by averaging features of the five known attacks form a region broad enough that the four held-out attack types fall inside it, and that the weak classifiers' out-of-distribution accuracy stays above roughly 40–50%; if either fails, the reported zero-shot accuracy would not transfer.

Editorial extensions

If this is right

  • A zero-day attack can be flagged on its first packet with 85–100% accuracy, giving operators a chance to act before the attack propagates.
  • Because normal traffic is classified with 100% accuracy, the approach can be deployed without flooding operators with false alarms.
  • The detector adapts to an ongoing attack within a few in-context examples (max-shot accuracy 94.9–100%), so it improves as the attack continues.
  • With a mix of 5% ground-truth and 95% weak-classifier label distributions during training, the distributional transformer avoids the hyperparameter sensitivity of the hard-label variant.
  • The reported per-sample inference time of 2.27 ms on a GPU with TensorRT fits inside the 3 ms IEC-61850 transmission budget, suggesting real-time deployment is plausible.

Reading between the lines

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

  • If real zero-day attacks use packet structures far from linear mixtures of the five known classes, the reported zero-shot accuracy is not guaranteed to transfer; the paper does not test that distribution-shift regime.
  • A straightforward extension would be to apply the same weak-label plus distributional-transformer pipeline to other industrial control protocols, such as Modbus or DNP3, to see whether the in-context generalization is specific to IEC-61850 traffic or a broader property of transformer sequence models.
  • The sensitivity analysis implies a practical rule of thumb: maintain weak-classifier accuracy above roughly 40–50% on out-of-distribution traffic, or the transformer's detection advantage shrinks and may disappear.
  • An adversarial implication the authors do not discuss: an attacker who sends a burst of similar attack packets actually trains the detector's in-context signal, so a stealthier adversary would vary features across packets to avoid giving the transformer coherent in-context evidence.
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

3 major / 6 minor

Summary. The paper proposes an in-context learning (ICL) approach to zero-day attack detection in IEC-61850 digital substations. The authors train GPT-2-based transformer variants on synthetic attack classes generated by a multi-mixing procedure, condition the model on weak-classifier hard labels (simple transformer, TF) or full class distributions (distributional transformer, DTF), and evaluate on held-out attack types from the ERENO dataset. They report 85.8–100% zero-shot and 94.9–100% max-shot OOD detection accuracy, 100% accuracy on normal traffic, and a deployment latency analysis showing compatibility with IEC-61850 timing constraints. The central claim is that MDTF detects zero-day attacks above 85% accuracy while existing baselines fail.

Significance. If the results are unbiased, the paper provides a practically relevant demonstration of transformers' in-context learning for intrusion detection, with a useful deployment-oriented timing model and a sensitivity analysis relating weak-classifier quality to final detection accuracy. The use of a real IEC-61850 dataset, comparison with several standard classifiers, and the explicit treatment of weak labels are strengths. However, the headline numerical claims are weakened by the absence of a validation split for hyperparameter selection, and the abstract's statement that 'state-of-the-art baselines fail' is an overstatement of what Table 1 shows. The work is therefore an interesting empirical contribution whose main quantitative claims require re-evaluation under a cleaner protocol.

major comments (3)
  1. [Sections 2.1, 2.2, Table 1] The zero-day detection numbers in Table 1 are likely optimistically biased because the key hyperparameters were selected using the same OOD attacks that later constitute the test set. Section 2.1 reports that Ksyn=700 'yielded the best performance' based on Figure 3's curves, which are computed on exactly the four OOD attack types (Poisoned High Rate, Masquerade Fake Fault, Masquerade Fake Normal, Random Replay). Section 2.2 similarly states that mixing ratios 60/40 for TF and 95/5 for DTF 'yielded the highest attack detection accuracy during testing,' with no validation split described. Since these same OOD attacks are used in Table 1, the reported accuracies are in-sample model-selection estimates rather than unbiased zero-day generalization estimates. The authors should either introduce a validation split for hyperparameter selection or transparently report Table 1 as development-set results and provide a separate validation protocol for the final claim.
  2. [Abstract and Table 1] The abstract's claim that 'the existing state-of-the-art baselines fail' is not supported by Table 1. Logistic Regression, Decision Tree, Random Forest, SVM, RNN, and LSTM achieve 1.000 accuracy on Poisoned High Rate, Masquerade Fake Fault, and Random Replay; their failure is concentrated on Masquerade Fake Normal (and Random Replay for Naive Bayes, DNN, CNN, and hard voting). The contribution is therefore better described as improving detection on the hardest OOD attack types, not as a blanket failure of all baselines. The abstract and Section 2.4 should be revised to state this more precisely.
  3. [Section 4.1 and Table 1] The external validity of the 'zero-day' claim rests on an unverified assumption that held-out attacks lie near the synthetic manifold generated by multi-mixing. Algorithm 1 constructs new classes as equal-weight arithmetic means of subsets of the five known attack classes, so the four ERENO OOD attacks used in Table 1 are tested under the assumption that they are plausible points on or near this synthetic manifold. The paper provides no analysis quantifying this proximity (e.g., distances in feature space or a visualization) and no evaluation on attacks generated by a different process. The zero-day generalization claim should be scoped accordingly, or such an analysis should be added.
minor comments (6)
  1. [Section 2.5 and Table 2] The text states that GPU TensorRT yields Taverage = 2.44 ms for BS=1, but Table 2 reports 0.00227 ± 0.00030 s, which is 2.27 ms. Please correct this inconsistency.
  2. [Section 4.1] Equation (1) defines Anew as a weighted sum with coefficients α_k, but Algorithm 1 uses equal weights of 1/len(attack_mixture). The relationship between α_k and the algorithm's uniform averaging should be clarified.
  3. [Section 3] There is a typo in the Discussion: 'time senstive' should be 'time-sensitive'.
  4. [General] No code or data availability statement is provided. Since the ERENO dataset is public, a link to the code would significantly improve reproducibility.
  5. [Table 1] The DNN row contains the entry '0' for Poisoned High Rate; format it as 0.000 for consistency with the other entries.
  6. [Section 2.3] The sensitivity analysis trains synthetic weak classifiers with some OOD attacks included as in-distribution to vary their accuracy. This is a useful simulation, but it should be explicitly labeled as a favorable-condition ablation, because deployed weak classifiers in a true zero-day scenario would not have seen the OOD attack classes.

Circularity Check

2 steps flagged · score 5.0 of 10

Reported zero-day accuracies are partly in-sample: Ksyn and weak/ground-truth mixing ratios are selected on the same OOD test attacks that Table 1 reports, so the headline >85% figures are best-of-grid values rather than unbiased predictions.

  1. fitted input called prediction [Section 2.1 (The Impact of Training Data Diversity)]
    "As the training dataset with 700 attack classes yielded the best performance, the following experiments are reported for this case only."

    Ksyn is chosen as the value with 'the best performance' on the OOD attacks, and the same OOD attack set is then used to report Table 1's zero-day accuracies. No validation split is described, so selecting Ksyn from {100, 300, 500, 700} on the four OOD attack types makes the reported 85%+ numbers the maximum over the searched configurations. The prediction is thus fitted to the target set before being presented as zero-day generalization.

  2. fitted input called prediction [Section 2.2 (Comparative Analysis of Label Choices During Training)]
    "we discovered that for the TF models, a training ratio of 60% weak classifier labels and 40% ground-truth labels yielded the highest attack detection accuracy during testing. In contrast, for the DTF model, we achieved the best performance with a minimal ground-truth label ratio of 5%."

    The mixing ratios 60/40 (TF) and 95/5 (DTF) are selected by maximizing accuracy 'during testing,' and the OOD attacks being tested are the same four attacks tabulated in Table 1. The reported MDTF/MTF zero-day accuracies are therefore the best over the searched ratios, so the central 'prediction' contains a test-set-selection component. This is a fitted-input-called-prediction pattern rather than an unbiased out-of-sample measurement.

full rationale

The paper's core derivation is empirical rather than formal: multi-mixing generates synthetic training classes as arithmetic averages of known attack features, weak classifiers produce labels/distributions, and a GPT-2 transformer performs in-context prediction. The main circularity concern is hyperparameter selection on the OOD test set: Section 2.1 reports Ksyn=700 as 'the best performance' and Section 2.2 reports mixing ratios that 'yielded the highest attack detection accuracy during testing,' with no validation split described. Because Table 1's headline zero-day accuracies come from the same OOD attacks used for these selections, the >85% figures are partially in-sample maxima rather than clean predictions. This is not a full 'equivalence by construction' because the transformer's ICL transformation of weak labels is independent content, and the model does outperform hard-voting weak classifiers on the harder OOD attacks. Self-citation [21] is present but not load-bearing: multi-mixing is a method inherited from the authors' prior work, not an external theorem invoked to force the result. Separately, the abstract's claim that 'state-of-the-art baselines fail' overstates Table 1, where LR, DT, RF, SVM, RNN, and LSTM reach 1.000 on three of the four OOD attacks; that is a reporting overclaim, not circularity. Overall, the central claim is partially compromised by test-set-tuned hyperparameters, giving a circularity score of 5.

Assumptions & free parameters 5 free parameters · 5 assumptions · 1 invented entities

The central claim depends on the synthetic mixture manifold being representative of unseen attacks, and on the ERENO dataset's attacks being adequate zero-day proxies. Several hyperparameters, including Ksyn and both mixing ratios, were selected based on OOD test performance, which increases the circularity burden. Weak classifier behavior and deployment timing assumptions add further domain-specific premises that are acknowledged but not independently verified.

free parameters (5)
  • Ksyn (number of synthetic classes) = 700
    Section 2.1 says the training dataset with 700 attack classes yielded the best performance, so this is selected on OOD test results.
  • TF mixing ratio (weak:ground-truth) = 60:40
    Section 2.2 states this ratio yielded the highest attack detection accuracy during testing.
  • DTF mixing ratio (weak:ground-truth) = 95:5
    Section 2.2 states that 5% ground truth was best for DTF and that higher ratios caused instability; chosen by test performance.
  • In-context sample size N = 11
    Set in Section 2 without reported sensitivity analysis; controls the amount of context the transformer sees.
  • Downsampling factors DSV and DGOOSE = DSV=1/12, DGOOSE=1
    Chosen in Section 2.5 to satisfy the 3 ms IEC-61850 timing constraint; detection accuracy under downsampling is not evaluated.
assumptions (5)
  • domain assumption Linear mixtures (averages) of features from known attack classes create synthetic classes that capture the diversity needed for in-context learning and are representative of future zero-day attacks.
    Multi-mixing in Section 4.1 builds training classes from arithmetic means of known attack data; no evidence is given that real zero-day attacks lie in this linear hull.
  • domain assumption The ERENO-IEC-61850 dataset's held-out attack classes (Poisoned High Rate, Masquerade Fake Fault, Masquerade Fake Normal, Random Replay) are adequate proxies for zero-day attacks in real substations.
    Evaluation uses only this public dataset; no deployment data or other testbeds are included.
  • domain assumption Weak classifiers' predictions on out-of-distribution packets carry enough signal for the transformer to condition on, even when the classifiers are inaccurate.
    The sensitivity analysis covers weak classifier accuracy from about 0.4 to 1.0, and the authors state performance may vary below 40-50% in the Limitations section.
  • domain assumption Deployment timing assumptions: Tpre is negligible, Tb dominates Ta+Tc, and the network controller can interrupt immediately upon detection.
    Section 2.5 lists these simplifying assumptions for the 3 ms deadline; they are not validated in hardware-in-the-loop.
  • domain assumption Transformer in-context learning emerges from task diversity in synthetic classification tasks, as demonstrated in prior language-model literature.
    The paper relies on [32-36] to justify that adding synthetic classes improves ICL; this is an empirical property imported from external work.
invented entities (1)
  • Synthetic multi-mixed attack classes
    purpose: Expand the training set from 5 original attack classes to up to 700 classes so the transformer can learn diverse attack patterns.
    These classes are arithmetic mixtures of known attack features; they are not validated against real attack traffic, so they are a postulated training construct rather than empirically verified attack patterns.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Detecting Zero-Day Attacks in Digital Substations via In-Context Learning." pith.science (2026). https://pith.science/paper/DVFDOVCN

@misc{pith2026250116453,
  author       = {Pith},
  title        = {Pith review of: Detecting Zero-Day Attacks in Digital Substations via In-Context Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DVFDOVCN}},
  note         = {Machine review of arXiv:2501.16453}
}
abstract

The occurrences of cyber attacks on the power grids have been increasing every year, with novel attack techniques emerging every year. In this paper, we address the critical challenge of detecting novel/zero-day attacks in digital substations that employ the IEC-61850 communication protocol. While many heuristic and machine learning (ML)-based methods have been proposed for attack detection in IEC-61850 digital substations, generalization to novel or zero-day attacks remains challenging. We propose an approach that leverages the in-context learning (ICL) capability of the transformer architecture, the fundamental building block of large language models. The ICL approach enables the model to detect zero-day attacks and learn from a few examples of that attack without explicit retraining. Our experiments on the IEC-61850 dataset demonstrate that the proposed method achieves more than $85\%$ detection accuracy on zero-day attacks while the existing state-of-the-art baselines fail. This work paves the way for building more secure and resilient digital substations of the future.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 33 canonical work pages

  1. [1]

    SANS Industrial Control Systems Security Blog 207 (2016) 16

    Assante, M.J.: Confirmation of a coordinated attack on the ukrainian power grid. SANS Industrial Control Systems Security Blog 207 (2016) 16

    work page 2016

  2. [2]

    Electricity Information Sharing and Analysis Center (E-ISAC) 388(1-29), 3 (2016)

    Case, D.U.: Analysis of the cyber attack on the ukrainian power grid. Electricity Information Sharing and Analysis Center (E-ISAC) 388(1-29), 3 (2016)

    work page 2016

  3. [3]

    http://www.statista.com/statistics/680953/global-malwarevolume/

    Global new malware volume (2020) statista. http://www.statista.com/statistics/680953/global-malwarevolume/

    work page 2020

  4. [4]

    In: 2017 IEEE International Conference on Smart Grid Communications (SmartGridComm), pp

    Duman, O., Zhang, M., Wang, L., Debbabi, M.: Measuring the security posture of iec 61850 substations with redundancy against zero day attacks. In: 2017 IEEE International Conference on Smart Grid Communications (SmartGridComm), pp. 108–114 (2017). IEEE

    work page 2017

  5. [5]

    IEEE Communications Magazine 61(6), 28–34 (2023)

    Lozano, J.C., Koneru, K., Ortiz, N., Cardenas, A.A.: Digital substations and iec 61850: A primer. IEEE Communications Magazine 61(6), 28–34 (2023)

    work page 2023

  6. [6]

    In: 2024 IEEE Power & Energy Society General Meeting (PESGM), pp

    Girdhar, M., Hong, J., Su, W., Herath, A., Liu, C.-C.: Sdn-based dynamic cybersecurity framework of iec-61850 communications in smart grid. In: 2024 IEEE Power & Energy Society General Meeting (PESGM), pp. 1–5 (2024). https://doi.org/10.1109/PESGM51994.2024.10688802

    arXiv 2024

  7. [7]

    In: 2012 IEEE PES Innovative Smart Grid Technologies (ISGT), pp

    Stefanov, A., Liu, C.-C.: Cyber-power system security in a smart grid environ- ment. In: 2012 IEEE PES Innovative Smart Grid Technologies (ISGT), pp. 1–3 (2012). IEEE

    work page 2012

  8. [8]

    IEC Std 61850 (2013)

    Commission, I.E., et al.: Communication networks and systems for power utility automation. IEC Std 61850 (2013)

    work page 2013

Show all 39 references
  1. [9]

    Journal of Electrical Systems and Information Technology 5(3), 468–483 (2018)

    Otuoze, A.O.e.a.: Smart grids security challenges: Classification by sources of threats. Journal of Electrical Systems and Information Technology 5(3), 468–483 (2018)

    2018

  2. [10]

    IEEE Transactions on Smart Grid 5(4), 1643–1653 (2014)

    Hong, J., Liu, C.-C., Govindarasu, M.: Integrated anomaly detection for cyber security of the substations. IEEE Transactions on Smart Grid 5(4), 1643–1653 (2014)

    2014

  3. [11]

    ACM Transactions on Cyber-Physical Systems 7(2), 1–31 (2023)

    Sahani, N., Zhu, R., Cho, J.-H., Liu, C.-C.: Machine learning-based intru- sion detection for smart grid computing: A survey. ACM Transactions on Cyber-Physical Systems 7(2), 1–31 (2023)

    2023

  4. [12]

    https://arxiv.org/abs/2411.07419

    Park, K., Girdhar, M., Hong, J., Su, W., Herath, A., Liu, C.-C.: Machine Learning Based Cyber System Restoration for IEC 61850 Based Digital Substations (2024). https://arxiv.org/abs/2411.07419

    2024 arXiv

  5. [13]

    In: ISGT 2014, pp

    Hong, J., Liu, C.-C., Govindarasu, M.: Detection of cyber intrusions using network-based multicast messages for substation automation. In: ISGT 2014, pp. 1–5 (2014). IEEE

    2014

  6. [14]

    In: 2020 IEEE PES Innovative Smart Grid Technologies Europe (ISGT-Europe) (2020)

    Rajkumar, V.S.e.a.: Cyber attacks on power system automation and protection 17 and impact analysis. In: 2020 IEEE PES Innovative Smart Grid Technologies Europe (ISGT-Europe) (2020). IEEE

    2020

  7. [15]

    In: 2016 IEEE Power and Energy Society General Meeting (PESGM), pp

    Yang, Y.e.a.: Intrusion detection system for iec 61850 based smart substations. In: 2016 IEEE Power and Energy Society General Meeting (PESGM), pp. 1–5 (2016). IEEE

    2016

  8. [16]

    Energies 12(19), 3731 (2019)

    El Hariri, M.e.a.: The iec 61850 sampled measured values protocol: Analysis, threat identification, and feasibility of using nn forecasters to detect spoofed packets. Energies 12(19), 3731 (2019)

    2019

  9. [17]

    Ieee Access 9, 56486– 56495 (2021)

    et.al., U.: Artificial intelligence based intrusion detection system for iec 61850 sampled values under symmetric and asymmetric faults. Ieee Access 9, 56486– 56495 (2021)

    2021

  10. [18]

    Advances in neural information processing systems 30 (2017)

    et.al., V.: Attention is all you need. Advances in neural information processing systems 30 (2017)

    2017

  11. [19]

    arXiv preprint arXiv:2005.14165 (2020)

    Brown, T.B.: Language models are few-shot learners. arXiv preprint arXiv:2005.14165 (2020)

    2020 arXiv

  12. [21]

    In: 2024 IEEE International Confer- ence on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm), pp

    Manzoor, F., Khattar, V., Liu, C.-C., Jin, M.: Zero-day attack detection in digital substations using in-context learning. In: 2024 IEEE International Confer- ence on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm), pp. 220–225 (2024). IEEE

    2024

  13. [22]

    Journal of Big data 3, 1–40 (2016)

    Weiss, K., Khoshgoftaar, T.M., Wang, D.: A survey of transfer learning. Journal of Big data 3, 1–40 (2016)

    2016

  14. [23]

    arXiv preprint arXiv:2110.11334 (2021)

    et.al., Y.: Generalized out-of-distribution detection: A survey. arXiv preprint arXiv:2110.11334 (2021)

    2021 arXiv

  15. [24]

    https://arxiv.org/abs/2405.02989

    Abdeen, Z., Roy, P., Al-Tawaha, A., Jia, R., Freeman, L., Beling, P., Liu, C.-C., Sangiovanni-Vincentelli, A., Jin, M.: Defense against Joint Poison and Evasion Attacks: A Case Study of DERMS (2024). https://arxiv.org/abs/2405.02989

    2024 arXiv

  16. [25]

    In: The Eleventh International Conference on Learning Representations (2022)

    Khattar, V., Ding, Y., Sel, B., Lavaei, J., Jin, M.: A cmdp-within-online framework for meta-safe reinforcement learning. In: The Eleventh International Conference on Learning Representations (2022)

    2022

  17. [26]

    In: 2024 IEEE Interna- tional Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm), pp

    Abdeen, Z., Zhang, X., Gill, W., Jin, M.: Enhancing distribution system resilience: A first-order meta-rl algorithm for critical load restoration. In: 2024 IEEE Interna- tional Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm), p...

    2024

  18. [27]

    In: Matni, N., Morari, M., Pappas, G.J

    Sel, B., Tawaha, A., Ding, Y., Jia, R., Ji, B., Lavaei, J., Jin, M.: Learning-to-learn to guide random search: Derivative-free meta blackbox optimization on manifold. In: Matni, N., Morari, M., Pappas, G.J. (eds.) Proceedings of The 5th Annual Learning for Dynamics and Control...

    2023

  19. [28]

    National Science Review 5(1), 30–43 (2018)

    Zhang, Y., Yang, Q.: An overview of multi-task learning. National Science Review 5(1), 30–43 (2018)

    2018

  20. [29]

    IEEE Transactions on Dependable and Secure Computing (2023)

    Quincozes, S.E., Albuquerque, C., Passos, D., Moss´ e, D.: Ereno: A framework for generating realistic iec–61850 intrusion detection datasets for smart grids. IEEE Transactions on Dependable and Secure Computing (2023)

    2023

  21. [30]

    OpenAI blog 1(8), 9 (2019)

    et.al., A.R.: Language models are unsupervised multitask learners. OpenAI blog 1(8), 9 (2019)

    2019

  22. [31]

    Hou, D., Dolezilek, D.: Iec 61850 - what it can and cannot offer to traditional protection schemes. (2010). https://api.semanticscholar.org/CorpusID:43886997

    2010

  23. [32]

    Advances in Neural Information Processing Systems 36 (2024)

    Ravent´ os, A.e.a.: Pretraining task diversity and the emergence of non-bayesian in-context learning for regression. Advances in Neural Information Processing Systems 36 (2024)

    2024

  24. [33]

    arXiv preprint arXiv:2110.15943 (2021)

    Min, S., Lewis, M., Zettlemoyer, L., Hajishirzi, H.: Metaicl: Learning to learn in context. arXiv preprint arXiv:2110.15943 (2021)

    2021 arXiv

  25. [34]

    https://arxiv.org/abs/2212.06800

    Levy, I., Bogin, B., Berant, J.: Diverse Demonstrations Improve In-context Compositional Generalization (2023). https://arxiv.org/abs/2212.06800

    2023 arXiv

  26. [35]

    Zhang, H., Cisse, M., Dauphin, Y.N., Lopez-Paz, D.: mixup: Beyond Empirical Risk Minimization (2018)

    2018

  27. [36]

    Advances in Neural Information Processing Systems 35, 18878– 18891 (2022)

    Chan, S.e.a.: Data distributional properties drive emergent in-context learning in transformers. Advances in Neural Information Processing Systems 35, 18878– 18891 (2022)

    2022

  28. [37]

    Sensors 21(4), 1554 (2021) Nomenclature Model Parameters and Dimensions: I Number of training samples used to train the transformer model

    et.al., R.: Vulnerability and impact analysis of the iec 61850 goose protocol in the smart grid. Sensors 21(4), 1554 (2021) Nomenclature Model Parameters and Dimensions: I Number of training samples used to train the transformer model. Kori Number of original classes used to g...

    2021

  29. [38]

    Each prediction depends only on past observa- tions, naturally aligning with the temporal ordering of packet streams in digital substations

    Causal Attention Mechanism. Each prediction depends only on past observa- tions, naturally aligning with the temporal ordering of packet streams in digital substations

  30. [39]

    This facilitates handling variable-length input sequences without architectural changes, crucial for evolving network traffic patterns

    Parameter-Sharing Through T ransformer Blocks. This facilitates handling variable-length input sequences without architectural changes, crucial for evolving network traffic patterns

  31. [40]

    By omitting an encoder stage, the model can efficiently process real-time data while preserving strong pattern- recognition capabilities

    Streamlined Decoder-Only Inference. By omitting an encoder stage, the model can efficiently process real-time data while preserving strong pattern- recognition capabilities. Although these architectural choices draw from GPT-2’s language modeling advances, they generalize to i...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.