pith. sign in

arxiv: 2409.07609 · v2 · submitted 2024-09-11 · 💻 cs.CR · cs.CV· cs.LG· stat.AP

Survival of the Cheapest: Cost-Aware Hardware Adaptation for Adversarial Robustness

Charles Meyers , Mohammad Reza Saleh Sedghpour , Tommy L\"ofstedt , Erik Elmroth This is my paper

Pith reviewed 2026-05-23 21:12 UTC · model grok-4.3

classification 💻 cs.CR cs.CVcs.LGstat.AP
keywords adversarial robustnesshardware selectioncost optimizationsurvival analysismachine learning deploymentinference latencyGPU comparison
0
0 comments X

The pith

Cheaper GPUs can increase a model's survival time against attacks by 20 percent while cutting costs by 75 percent.

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

The paper introduces a quantitative approach that models how hardware choices and training settings affect how long a machine learning model resists adversarial attacks. It treats survival time under attack as the key outcome and shows that this time can be predicted and optimized to favor lower-cost options. Experiments on three GPU types confirm that a less expensive card outperforms a pricier one on robustness while using far fewer resources. The analysis identifies inference latency as the strongest signal of robustness, ahead of training time or hardware type itself.

Core claim

The paper establishes that accelerated failure time models applied to adversarial survival time can quantify the impact of hardware selection, batch size, epochs, and validation accuracy, with results showing the Nvidia L4 delivering 20 percent longer survival time at 75 percent lower cost than the V100 and inference latency emerging as the dominant predictor over training duration or hardware configuration.

What carries the argument

Accelerated failure time models that treat adversarial survival time as the outcome variable driven by hardware, batch size, epochs, and validation accuracy.

If this is right

  • Hardware and hyperparameter choices can be ranked directly by predicted survival time to favor cheaper configurations that still raise robustness.
  • Inference latency becomes the primary metric for deciding where to run models, shifting attention from training hardware to deployment speed.
  • Continuous monitoring of cost, robustness, and latency can drive automatic selection of new hardware or settings as conditions change.

Where Pith is reading between the lines

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

  • The same survival modeling could be tested on non-GPU accelerators or CPU-only setups to see whether the cost-robustness pattern holds.
  • Optimizing explicitly for inference latency might yield robustness gains even without changing hardware, a direct extension worth checking.
  • The approach could apply to measuring resilience against other threats such as data poisoning if survival time is redefined accordingly.

Load-bearing premise

Survival time until a model fails under adversarial attack is a valid and sufficient measure of robustness, and the models accurately capture how hardware and other factors change that time.

What would settle it

Run the same adversarial attacks on models trained and deployed on L4 versus V100 GPUs and check whether the measured failure times match the survival times predicted by the models.

read the original abstract

Deploying adversarially robust machine learning systems requires continuous trade-offs between robustness, cost, and latency. We present an autonomic decision-support framework providing a quantitative foundation for adaptive hardware selection and hyper-parameter tuning in cloud-native deep learning. The framework applies accelerated failure time (AFT) models to quantify the effect of hardware choice, batch size, epochs, and validation accuracy on model survival time. This framework can be naturally integrated into an autonomic control loop (monitor--analyse--plan--execute, MAPE-K), where system metrics such as cost, robustness, and latency are continuously evaluated and used to adapt model configurations and hardware selection. Experiments across three GPU architectures confirm the framework is both sound and cost-effective: the Nvidia L4 yields a 20% increase in adversarial survival time while costing 75% less than the V100, demonstrating that expensive hardware does not necessarily improve robustness. The analysis further reveals that model inference latency is a stronger predictor of adversarial robustness than training time or hardware configuration.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 0 minor

Summary. The paper proposes an autonomic decision-support framework that applies accelerated failure time (AFT) models to quantify how hardware choice, batch size, epochs, and validation accuracy affect 'adversarial survival time' as a robustness measure. It claims experiments across three GPU architectures demonstrate the framework is sound and cost-effective, with the Nvidia L4 yielding a 20% increase in adversarial survival time at 75% lower cost than the V100; inference latency is reported as a stronger predictor of robustness than training time or hardware configuration. The framework is positioned for integration into MAPE-K control loops for adaptive hardware and hyper-parameter selection.

Significance. If the adversarial survival time metric were shown to correlate with established robustness measures, the work could provide a practical, quantitative basis for cost-aware hardware adaptation in robust ML deployments, with the latency-predictor finding offering actionable system-design guidance. The cost-effectiveness demonstration (L4 vs. V100) would then carry weight for cloud-native settings. At present the significance is constrained by the unvalidated metric.

major comments (3)
  1. [Abstract] Abstract: The central claim that the L4 hardware yields a 20% increase in adversarial survival time (hence better robustness) at 75% lower cost than the V100 rests on survival time being a meaningful proxy for adversarial robustness. The manuscript supplies no evidence that this metric correlates with or substitutes for standard measures such as robust accuracy under PGD/FGSM attacks or attack success rates; without this link the hardware-adaptation conclusions cannot be interpreted as robustness improvements.
  2. [Abstract] Abstract: The AFT models are asserted to quantify the effects of hardware, batch size, epochs, and validation accuracy on survival time and to identify inference latency as the strongest predictor, yet no details are provided on the AFT functional form, the definition of the failure event, censoring treatment, coefficient estimation, or any goodness-of-fit or significance tests. These omissions make it impossible to verify the reported 20% improvement or the predictor ranking.
  3. [Abstract] Abstract: The experimental claim of a 20% survival-time advantage for the L4 is presented without error bars, replication counts, statistical tests, or model specifications (architecture, attack parameters, training details). This prevents assessment of whether the difference is reliable or whether the three-GPU comparison supports the broader conclusion that expensive hardware does not necessarily improve robustness.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The comments highlight important aspects of metric validation and methodological transparency that we will address in revision. Below we respond point-by-point to the major comments.

read point-by-point responses

  1. Referee: [Abstract] The central claim that the L4 hardware yields a 20% increase in adversarial survival time at 75% lower cost than the V100 rests on survival time being a meaningful proxy for adversarial robustness. The manuscript supplies no evidence that this metric correlates with or substitutes for standard measures such as robust accuracy under PGD/FGSM attacks or attack success rates.

    Authors: We agree that demonstrating a link between adversarial survival time and established robustness metrics would strengthen interpretability. Adversarial survival time measures time-to-failure under sustained attack and is therefore conceptually related to robustness, yet we did not include an explicit correlation study. In the revised manuscript we will add a dedicated subsection reporting Pearson and Spearman correlations between survival time and robust accuracy (PGD, epsilon=0.03) across the evaluated configurations, together with scatter plots. revision: yes

  2. Referee: [Abstract] The AFT models are asserted to quantify the effects of hardware, batch size, epochs, and validation accuracy on survival time and to identify inference latency as the strongest predictor, yet no details are provided on the AFT functional form, the definition of the failure event, censoring treatment, coefficient estimation, or any goodness-of-fit or significance tests.

    Authors: The full manuscript (Section 3) specifies a log-normal AFT model, defines the failure event as the first epoch at which adversarial accuracy falls below 50 percent under the chosen attack, treats right-censored observations for runs that reached the maximum experiment duration, and reports coefficient estimates with p-values. To improve accessibility we will expand this section with the explicit survival function, the estimation procedure (maximum likelihood via survreg), and model diagnostics including the concordance index and residual plots. revision: partial

  3. Referee: [Abstract] The experimental claim of a 20% survival-time advantage for the L4 is presented without error bars, replication counts, statistical tests, or model specifications (architecture, attack parameters, training details).

    Authors: The 20 percent figure is the AFT-predicted multiplicative effect for the L4 versus V100 after controlling for batch size and epochs. We will augment the results section with bootstrap-derived 95 percent confidence intervals on the survival-time ratios, the number of independent training runs per hardware-batch combination (five), and the results of likelihood-ratio tests comparing nested AFT models. Attack parameters (PGD, 20 iterations, epsilon=0.03) and model architecture (ResNet-18) are already stated in the experimental setup; we will ensure they are also summarized in the abstract revision. revision: yes

Circularity Check

0 steps flagged

No significant circularity; claims rest on empirical AFT fitting to measured data

full rationale

The paper fits AFT models to experimentally measured adversarial survival times across GPU architectures, batch sizes, epochs, and validation accuracies, then reports direct comparisons (e.g., L4 vs V100 survival time and cost). No step reduces a claimed prediction or result to its own fitted inputs by construction, nor relies on self-citation chains, uniqueness theorems from the authors, or smuggled ansatzes. Survival time is treated as an observed quantity whose effects are quantified by regression; the derivation chain remains independent of the target conclusions.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Only the abstract is available, so the ledger reflects elements explicitly named in the abstract; AFT model parameters are presumed fitted to data but not enumerated.

axioms (1)
  • domain assumption Accelerated failure time models are suitable for modeling adversarial model robustness as a survival-time outcome dependent on hardware and training variables.
    The entire framework rests on treating robustness via AFT survival analysis.
invented entities (1)
  • adversarial survival time no independent evidence
    purpose: To serve as a quantitative metric for how long a model remains robust under attack as a function of configuration variables.
    This metric is introduced as the target variable for the AFT models and the basis for hardware selection decisions.

pith-pipeline@v0.9.0 · 5720 in / 1398 out tokens · 28434 ms · 2026-05-23T21:12:17.919000+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

76 extracted references · 76 canonical work pages · 10 internal anchors

  1. [1]

    Radiographics 37(2), 505–515 (2017) 23

    Erickson, B.J., Korfiatis, P., Akkus, Z., Kline, T.L.: Machine learning for medical imaging. Radiographics 37(2), 505–515 (2017) 23

    work page 2017

  2. [2]

    In: 2018 Aviation Technology, Integration, and Operations Conference, p

    Maheshwari, A., Davendralingam, N., DeLaurentis, D.A.: A comparative study of machine learning techniques for aviation applications. In: 2018 Aviation Technology, Integration, and Operations Conference, p. 3980 (2018)

    work page 2018

  3. [3]

    In: 31st USENIX Security Symposium (USENIX Security 22), pp

    Arp, D., Quiring, E., Pendlebury, F., Warnecke, A., Pierazzi, F., Wressnegger, C., Cavallaro, L., Rieck, K.: Dos and don’ts of machine learning in computer security. In: 31st USENIX Security Symposium (USENIX Security 22), pp. 3971–3988 (2022)

    work page 2022

  4. [4]

    IEEE Transactions on Systems, Man, and Cybernetics: Systems 47(4), 682–692 (2016)

    Mery, D., Svec, E., Arias, M., Riffo, V., Saavedra, J.M., Banerjee, S.: Modern com- puter vision techniques for x-ray testing in baggage inspection. IEEE Transactions on Systems, Man, and Cybernetics: Systems 47(4), 682–692 (2016)

    work page 2016

  5. [5]

    International journal of environmental research and public health 19(17), 10594 (2022)

    Travaini, G.V., Pacchioni, F., Bellumore, S., Bosia, M., De Micco, F.: Machine learning and criminal justice: A systematic review of advanced methodology for recidivism risk prediction. International journal of environmental research and public health 19(17), 10594 (2022)

    work page 2022

  6. [6]

    End to End Learning for Self-Driving Cars

    Bojarski, M., Del Testa, D., Dworakowski, D., Firner, B., Flepp, B., Goyal, P., Jackel, L.D., Monfort, M., Muller, U., Zhang, J., et al.: End to end learning for self-driving cars. arXiv preprint arXiv:1604.07316 (2016)

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  7. [7]

    IEEE transactions on neural networks 10(5), 988–999 (1999)

    Vapnik, V.N.: An overview of statistical learning theory. IEEE transactions on neural networks 10(5), 988–999 (1999)

    work page 1999

  8. [8]

    Cambridge university press, Cambridge, UK (2014)

    Shalev-Shwartz, S., Ben-David, S.: Understanding Machine Learning: From Theory to Algorithms. Cambridge university press, Cambridge, UK (2014)

    work page 2014

  9. [9]

    arXiv:2109.05472 (2021)

    Desislavov, R., Mart´ ınez-Plumed, F., Hern´ andez-Orallo, J.: Compute and energy consumption trends in deep learning inference. arXiv:2109.05472 (2021)

    work page arXiv 2021

  10. [10]

    Computer Methods and Programs in Biomedicine 213, 106504 (2022)

    Bailly, A., Blanc, C., Francis, ´E., Guillotin, T., Jamal, F., Wakim, B., Roy, P.: Effects of dataset size and interactions on the prediction performance of logis- tic regression and deep learning models. Computer Methods and Programs in Biomedicine 213, 106504 (2022)

    work page 2022

  11. [11]

    In: Proceedings of the IEEE International Conference on Computer Vision, pp

    Sun, C., Shrivastava, A., Singh, S., Gupta, A.: Revisiting unreasonable effec- tiveness of data in deep learning era. In: Proceedings of the IEEE International Conference on Computer Vision, pp. 843–852 (2017)

    work page 2017

  12. [12]

    Artificial Intelligence Review (2023)

    Meyers, C., L¨ ofstedt, T., Elmroth, E.: Safety-critical computer vision: An empiri- cal survey of adversarial evasion attacks and defenses on computer vision systems. Artificial Intelligence Review (2023)

    work page 2023

  13. [13]

    tools fueled a 34% spike in Microsoft’s water consumption, and one city with its data centers is concerned about the effect on residential supply

    O’Brien, M., Fingerhut, H., Press, T.A.: A.I. tools fueled a 34% spike in Microsoft’s water consumption, and one city with its data centers is concerned about the effect on residential supply. Fortune (2023). https://fortune.com/2023/ 24 09/09/ai-chatgpt-usage-fuels-spike-in-microsoft-water-consumption/

    work page 2023

  14. [14]

    The History Began from AlexNet: A Comprehensive Survey on Deep Learning Approaches

    Alom, M.Z., Taha, T.M., Yakopcic, C., Westberg, S., Sidike, P., Nasrin, M.S., Van Esesn, B.C., Awwal, A.A.S., Asari, V.K.: The history began from alexnet: A comprehensive survey on deep learning approaches. arXiv preprint arXiv:1803.01164 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  15. [15]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp

    He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recogni- tion. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 770–778 (2016)

    work page 2016

  16. [16]

    An Empirical Study of Mamba-based Language Models

    Waleffe, R., Byeon, W., Riach, D., Norick, B., Korthikanti, V., Dao, T., Gu, A., Hatamizadeh, A., Singh, S., Narayanan, D., et al.: An empirical study of mamba-based language models. arXiv preprint arXiv:2406.07887 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  17. [17]

    International Elec- trotechnical Commission (2010)

    Commission, I.E.: IEC 61508 Safety and Functional Safety. International Elec- trotechnical Commission (2010)

    work page 2010

  18. [18]

    https://www.iso.org/standard/43464.html (2018)

    International Standards Organization: ISO 26262-1:2011, Road vehicles — Func- tional safety. https://www.iso.org/standard/43464.html (2018)

    work page 2011

  19. [19]

    In: 2011 IEEE Long Island Systems, Applications and Technology Conference, pp

    Axelrod, C.W.: Applying lessons from safety-critical systems to security-critical software. In: 2011 IEEE Long Island Systems, Applications and Technology Conference, pp. 1–6 (2011). IEEE

    work page 2011

  20. [20]

    Safety Science 77, 133–142 (2015)

    Acharyulu, P.S., Seetharamaiah, P.: A framework for safety automation of safety- critical systems operations. Safety Science 77, 133–142 (2015)

    work page 2015

  21. [21]

    Adversarial Patch

    Brown, T.B., Man´ e, D., Roy, A., Abadi, M., Gilmer, J.: Adversarial patch. arXiv:1712.09665 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  22. [22]

    In: 2017 IEEE Symposium on Security and Privacy (SP), pp

    Carlini, N., Wagner, D.: Towards evaluating the robustness of neural networks. In: 2017 IEEE Symposium on Security and Privacy (SP), pp. 39–57 (2017)

    work page 2017

  23. [23]

    In: International Conference on Machine Learning, pp

    Croce, F., Hein, M.: Reliable evaluation of adversarial robustness with an ensem- ble of diverse parameter-free attacks. In: International Conference on Machine Learning, pp. 2206–2216 (2020). PMLR

    work page 2020

  24. [24]

    In: IEEE Symposium on Security and Privacy (SP), pp

    Chen, J., Jordan, M.I., Wainwright, M.J.: HopSkipJumpAttack: A query-efficient decision-based attack. In: IEEE Symposium on Security and Privacy (SP), pp. 1277–1294 (2020). IEEE

    work page 2020

  25. [26]

    Papernot, N., McDaniel, P

    Nicolae, M.I., Sinn, M., Tran, M.N., Buesser, B., Rawat, A., Wistuba, M., Zantedeschi, V., Baracaldo, N., Chen, B., Ludwig, H., Molloy, I., Edwards, B.: Adversarial robustness toolbox v1.2.0. CoRR 1807.01069 (2018) 25

    work page arXiv 2018

  26. [27]

    arXiv preprint arXiv:2403.00787 (2024)

    Panchal, D., Verma, P., Baran, I., Musgrove, D., Lu, D.: Reusable mlops: Reusable deployment, reusable infrastructure and hot-swappable machine learning models and services. arXiv preprint arXiv:2403.00787 (2024)

    work page arXiv 2024

  27. [28]

    In: 2017 IEEE International Conference on Software Architecture Workshops (ICSAW), pp

    Hasselbring, W., Steinacker, G.: Microservice architectures for scalability, agility and reliability in e-commerce. In: 2017 IEEE International Conference on Software Architecture Workshops (ICSAW), pp. 243–246 (2017). IEEE

    work page 2017

  28. [29]

    IEEE Transactions on Cloud Computing (2022)

    Zhou, R., Pang, J., Zhang, Q., Wu, C., Jiao, L., Zhong, Y., Li, Z.: Online scheduling algorithm for heterogeneous distributed machine learning jobs. IEEE Transactions on Cloud Computing (2022)

    work page 2022

  29. [30]

    Journal of Cloud Computing 12(1), 4 (2023)

    Singh, N., Hamid, Y., Juneja, S., Srivastava, G., Dhiman, G., Gadekallu, T.R., Shah, M.A.: Load balancing and service discovery using docker swarm for microservice based big data applications. Journal of Cloud Computing 12(1), 4 (2023)

    work page 2023

  30. [31]

    Github.com (2019)

    Github: Octoverse Projects. Github.com (2019). https://octoverse.github.com/ 2018/projects.html

    work page 2019

  31. [32]

    arXiv preprint arXiv:1907.10701 (2019)

    Wang, Y.E., Wei, G.-Y., Brooks, D.: Benchmarking tpu, gpu, and cpu platforms for deep learning. arXiv preprint arXiv:1907.10701 (2019)

    work page arXiv 1907

  32. [33]

    Github (2019)

    Kubernetes: Kubernetes–an open source system for managing containerized applications. Github (2019). https://github.com/kubernetes/kubernetes

    work page 2019

  33. [34]

    Github (2023)

    dvc.org: DVC- Data Version Control. Github (2023). https://github.com/ iterative/dvc.org

    work page 2023

  34. [35]

    Github (2019)

    Yadan, O.: Hydra - A framework for elegantly configuring complex applications. Github (2019). https://github.com/facebookresearch/hydra

    work page 2019

  35. [36]

    Advances in neural information processing systems 32 (2019)

    Cao, Y., Gu, Q.: Generalization bounds of stochastic gradient descent for wide and deep neural networks. Advances in neural information processing systems 32 (2019)

    work page 2019

  36. [37]

    Journal of Machine Learning Research 23(173), 1–65 (2022)

    Granziol, D., Zohren, S., Roberts, S.: Learning rates as a function of batch size: A random matrix theory approach to neural network training. Journal of Machine Learning Research 23(173), 1–65 (2022)

    work page 2022

  37. [38]

    In: Artificial Intelligence and Machine Learning for Multi-domain Operations Applications, vol

    Smith, L.N., Topin, N.: Super-convergence: Very fast training of neural networks using large learning rates. In: Artificial Intelligence and Machine Learning for Multi-domain Operations Applications, vol. 11006, pp. 369–386 (2019). SPIE

    work page 2019

  38. [39]

    In: International Conference on Cloud, Big Data and Trust, vol

    Sajid, M., Raza, Z.: Cloud computing: Issues & challenges. In: International Conference on Cloud, Big Data and Trust, vol. 20, pp. 13–15 (2013)

    work page 2013

  39. [40]

    Poisoning Attacks against Support Vector Machines

    Biggio, B., Nelson, B., Laskov, P.: Poisoning Attacks against Support Vector 26 Machines. arXiv:1206.6389 [cs, stat] (2013)

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  40. [41]

    In: Proceedings of the AAAI Conference on Artificial Intelligence, vol

    Saha, A., Subramanya, A., Pirsiavash, H.: Hidden trigger backdoor attacks. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 34, pp. 11957– 11965 (2020)

    work page 2020

  41. [42]

    Adversarial Attacks and Defences: A Survey

    Chakraborty, A., Alam, M., Dey, V., Chattopadhyay, A., Mukhopadhyay, D.: Adversarial attacks and defences: A survey. arXiv:1810.00069 [cs, stat] (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  42. [43]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Orekondy, T., Schiele, B., Fritz, M.: Knockoff nets: Stealing functionality of black- box models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 4954–4963 (2019)

    work page 2019

  43. [44]

    In: International Conference on Machine Learning, pp

    Choquette-Choo, C.A., Tramer, F., Carlini, N., Papernot, N.: Label-only mem- bership inference attacks. In: International Conference on Machine Learning, pp. 1964–1974 (2021). PMLR

    work page 1964

  44. [45]

    In: Proceedings of the 2021 ACM SIGSAC Conference on Computer and Communications Security, pp

    Li, Z., Zhang, Y.: Membership leakage in label-only exposures. In: Proceedings of the 2021 ACM SIGSAC Conference on Computer and Communications Security, pp. 880–895 (2021)

    work page 2021

  45. [46]

    PloS one 17(4), 0265723 (2022)

    Kotyan, S., Vargas, D.V.: Adversarial robustness assessment. PloS one 17(4), 0265723 (2022)

    work page 2022

  46. [47]

    Explaining and Harnessing Adversarial Examples

    Goodfellow, I.J., Shlens, J., Szegedy, C.: Explaining and harnessing adversarial examples. arXiv:1412.6572 (2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  47. [48]

    In: 29th USENIX Security Symposium (USENIX Security 20), pp

    Leurent, G., Peyrin, T.: SHA-1 is a shambles: First Chosen-Prefix collision on SHA-1 and application to the PGP web of trust. In: 29th USENIX Security Symposium (USENIX Security 20), pp. 1839–1856 (2020)

    work page 2020

  48. [49]

    Master’s thesis, St

    Kamal, P.: A study on the security of password hashing based on gpu based, password cracking using high-performance cloud computing. Master’s thesis, St. Cloud State. St. Cloud, Minnesota, USA. (2017)

    work page 2017

  49. [50]

    In: Machine Learning and Knowledge Discovery in Databases: European Conference, ECML PKDD 2013, Prague, Czech Republic, September 23-27, 2013, Proceedings, Part III 13, pp

    Biggio, B., Corona, I., Maiorca, D., Nelson, B., ˇSrndi´ c, N., Laskov, P., Giacinto, G., Roli, F.: Evasion attacks against machine learning at test time. In: Machine Learning and Knowledge Discovery in Databases: European Conference, ECML PKDD 2013, Prague, Czech Republic, September 23-27, 2013, Proceedings, Part III 13, pp. 387–402 (2013). Springer

    work page 2013

  50. [51]

    British journal of cancer 89(3), 431–436 (2003)

    Bradburn, M.J., Clark, T.G., Love, S.B., Altman, D.G.: Survival analysis part II: multivariate data analysis–an introduction to concepts and methods. British journal of cancer 89(3), 431–436 (2003)

    work page 2003

  51. [52]

    Springer, New York, NY, USA (2012) 27

    Kleinbaum, D.G., Klein, M.: Survival Analysis a Self-learning Text. Springer, New York, NY, USA (2012) 27

    work page 2012

  52. [53]

    arXiv preprint arXiv:2106.04525 (2021)

    Tan, J., Yang, J., Wu, S., Chen, G., Zhao, J.: A critical look at the current train/test split in machine learning. arXiv preprint arXiv:2106.04525 (2021)

    work page arXiv 2021

  53. [54]

    In: Proceedings of the 36th International Conference on Machine Learning

    Dohmatob, E.: Generalized No Free Lunch Theorem for Adversarial Robust- ness. In: Proceedings of the 36th International Conference on Machine Learning. PMLR, vol. 97 (2019)

    work page 2019

  54. [55]

    In: Accepted for International Conference on Machine Learning and Cybernetics (2023)

    Meyers, C., Reza, M., L¨ ofstedt, T., Elmroth, E.: A systematic approach to robust- ness modelling. In: Accepted for International Conference on Machine Learning and Cybernetics (2023)

    work page 2023

  55. [56]

    In: Pro- ceedings of the 2020 Genetic and Evolutionary Computation Conference, pp

    Ozaki, Y., Tanigaki, Y., Watanabe, S., Onishi, M.: Multiobjective tree-structured parzen estimator for computationally expensive optimization problems. In: Pro- ceedings of the 2020 Genetic and Evolutionary Computation Conference, pp. 533–541 (2020)

    work page 2020

  56. [57]

    In: Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp

    Akiba, T., Sano, S., Yanase, T., Ohta, T., Koyama, M.: Optuna: A next- generation hyperparameter optimization framework. In: Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp. 2623–2631 (2019)

    work page 2019

  57. [58]

    arXiv preprint arXiv:2304.11127 (2023)

    Watanabe, S.: Tree-structured parzen estimator: Understanding its algorithm components and their roles for better empirical performance. arXiv preprint arXiv:2304.11127 (2023)

    work page arXiv 2023

  58. [59]

    Statistics in Medicine 39(21), 2714–2742 (2020)

    Austin, P.C., Harrell Jr, F.E., Klaveren, D.: Graphical calibration curves and the integrated calibration index (ICI) for survival models. Statistics in Medicine 39(21), 2714–2742 (2020)

    work page 2020

  59. [60]

    Very Deep Convolutional Networks for Large-Scale Image Recognition

    Simonyan, K., Zisserman, A.: Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556 (2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  60. [61]

    In: 2023 IEEE 16th International Conference on Cloud Computing (CLOUD), pp

    Amaral, M., Chen, H., Chiba, T., Nakazawa, R., Choochotkaew, S., Lee, E.K., Eilam, T.: Kepler: A framework to calculate the energy consumption of con- tainerized applications. In: 2023 IEEE 16th International Conference on Cloud Computing (CLOUD), pp. 69–71 (2023)

    work page 2023

  61. [62]

    In: 2022 IEEE International Conference on Service- Oriented System Engineering (SOSE), pp

    Sedghpour, M.R.S., Townend, P.: Service mesh and ebpf-powered microservices: A survey and future directions. In: 2022 IEEE International Conference on Service- Oriented System Engineering (SOSE), pp. 176–184 (2022). https://doi.org/10. 1109/SOSE55356.2022.00027

    work page arXiv 2022

  62. [63]

    Multiobjective optimization: Interactive and evolutionary approaches, 373–404 (2008)

    Zitzler, E., Knowles, J., Thiele, L.: Quality assessment of pareto set approxi- mations. Multiobjective optimization: Interactive and evolutionary approaches, 373–404 (2008)

    work page 2008

  63. [64]

    In: International Conference on Tools and Algorithms for the Construction and Analysis of Systems, pp

    Legriel, J., Le Guernic, C., Cotton, S., Maler, O.: Approximating the pareto front 28 of multi-criteria optimization problems. In: International Conference on Tools and Algorithms for the Construction and Analysis of Systems, pp. 69–83 (2010). Springer

    work page 2010

  64. [65]

    IEEE Signal Processing Magazine 29(6), 141–142 (2012)

    Deng, L.: The MNIST database of handwritten digit images for machine learning research. IEEE Signal Processing Magazine 29(6), 141–142 (2012)

    work page 2012

  65. [66]

    Technical report, Toronto, ON, Canada (2009)

    Krizhevsky, A., Hinton, G., et al.: Learning multiple layers of features from tiny images. Technical report, Toronto, ON, Canada (2009)

    work page 2009

  66. [67]

    Towards Deep Learning Models Resistant to Adversarial Attacks

    Madry, A., Makelov, A., Schmidt, L., Tsipras, D., Vladu, A.: Towards deep learning models resistant to adversarial attacks. arXiv:1706.06083 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  67. [68]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp

    Moosavi-Dezfooli, S.-M., Fawzi, A., Frossard, P.: Deepfool: a simple and accurate method to fool deep neural networks. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2574–2582 (2016)

    work page 2016

  68. [69]

    Feature Squeezing: Detecting Adversarial Examples in Deep Neural Networks

    Xu, W., Evans, D., Qi, Y.: Feature squeezing: Detecting adversarial examples in deep neural networks. arXiv:1704.01155 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  69. [70]

    In: Proceedings of the 11th International Symposium on Highly Efficient Accelerators and Reconfigurable Technologies, pp

    Svedin, M., Chien, S.W., Chikafa, G., Jansson, N., Podobas, A.: Benchmarking the nvidia gpu lineage: From early k80 to modern a100 with asynchronous memory transfers. In: Proceedings of the 11th International Symposium on Highly Efficient Accelerators and Reconfigurable Technologies, pp. 1–6 (2021)

    work page 2021

  70. [71]

    In: 2018 IEEE/ACM Performance Modeling, Benchmarking and Simulation of High Performance Computer Systems (PMBS), pp

    Xu, R., Han, F., Ta, Q.: Deep learning at scale on nvidia v100 accelerators. In: 2018 IEEE/ACM Performance Modeling, Benchmarking and Simulation of High Performance Computer Systems (PMBS), pp. 23–32 (2018). IEEE

    work page 2018

  71. [72]

    Compute Engine: Virtual Machines (VMS)

    Google: GPU pricing. Compute Engine: Virtual Machines (VMS). google cloud. Google. https://cloud.google.com/compute/gpus-pricing

    work page

  72. [73]

    Compute Engine: Virtual Machines (VMS)

    Google: GPU pricing. Compute Engine: Virtual Machines (VMS). google cloud. Google. https://cloud.google.com/compute/vm-instance-pricing# accelerator-optimised

    work page

  73. [74]

    lifelines: survival analysis in Python.Journal of Open Source Software, 4(40):1317, 2019

    Davidson-Pilon, C.: lifelines: survival analysis in python. Journal of Open Source Software 4(40), 1317 (2019) https://doi.org/10.21105/joss.01317

    work page doi:10.21105/joss.01317 2019

  74. [75]

    IEEE Communications surveys & tutorials 18(1), 732–794 (2015)

    Dayarathna, M., Wen, Y., Fan, R.: Data center energy consumption modeling: A survey. IEEE Communications surveys & tutorials 18(1), 732–794 (2015)

    work page 2015

  75. [76]

    Statistics in medicine 19(4), 441–452 (2000)

    Schmoor, C., Sauerbrei, W., Schumacher, M.: Sample size considerations for the evaluation of prognostic factors in survival analysis. Statistics in medicine 19(4), 441–452 (2000)

    work page 2000

  76. [77]

    Controlled clinical trials 2(2), 93–113 (1981) 29

    Lachin, J.M.: Introduction to sample size determination and power analysis for clinical trials. Controlled clinical trials 2(2), 93–113 (1981) 29

    work page 1981