MODEL-AGNOSTIC META-LEARNING FOR RESILIENCE OPTIMIZATION OF ARTIFICIAL INTELLIGENCE SYSTEM
DOI:
https://doi.org/10.15588/1607-3274-2023-2-9Keywords:
Meta-learning, Evolutionary Strategies, Parameter-Efficient Transfer Learning, Robustness, Resilience, Adversarial Attacks, Faults Injection, Few-Shot LearningAbstract
Context. The problem of optimizing the resilience of artificial intelligence systems to destructive disturbances has not yet been fully solved and is quite relevant for safety-critical applications. The task of optimizing the resilience of an artificial intelligence system to disturbing influences is a high-level task in relation to efficiency optimization, which determines the prospects of using the ideas and methods of meta-learning to solve it. The object of current research is the process of meta-learning aimed at optimizing the resilience of an artificial intelligence system to destructive disturbances. The subjects of the study are architectural add-ons and the meta-learning method which optimize resilience to adversarial attacks, fault injection, and task changes.
Objective. Stated research goal is to develop an effective meta-learning method for optimizing the resilience of an artificial intelligence system to destructive disturbances.
Method. The resilience optimization is implemented by combining the ideas and methods of adversarial learning, fault-tolerant learning, model-agnostic meta-learning, few-shot learning, gradient optimization methods, and probabilistic gradient approximation strategies. The choice of architectural add-ons is based on parameter-efficient knowledge transfer designed to save resources and avoid the problem of catastrophic forgetting.
Results. A model-agnostic meta-learning method for optimizing the resilience of artificial intelligence systems based on gradient meta-updates or meta-updates using an evolutionary strategy has been developed. This method involves the use of tuner and metatuner blocks that perform parallel correction of the building blocks of a original deep neural network. The ability of the proposed approach to increase the efficiency of perturbation absorption and increase the integral resilience indicator of the artificial intelligence system is experimentally tested on the example of the image classification task. The experiments were conducted on a model with the ResNet-18 architecture, with an add-on in the form of tuners and meta-tuners with the Conv-Adapter architecture. In this case, CIFAR-10 is used as a base set on which the model was trained, and CIFAR-100 is used as a set for generating samples on which adaptation is performed using a few-shot learning scenarios. We compare the resilience of the artificial intelligence system after pre-training tuners and meta-tuners using the adversarial learning algorithm, the fault-tolerant learning algorithm, the conventional model-agnostic meta-learning algorithm, and the proposed meta-learning method for optimizing resilience. Also, the meta-learning algorithms with meta-gradient updating and meta-updating based on the evolutionary strategy are compared on the basis of the integral resilience indicator.
Conclusions. It has been experimentally confirmed that the proposed method provides a better resilience to random bit-flip injection compared to fault injection training by an average of 5%. Also, the proposed method provides a better resilience to Ladversarial evasion attacks compared to adversarial training by an average of 4.8%. In addition, an average 4.8% increase in the resilience to task changes is demonstrated compared to conventional fine-tuning of tuners. Moreover, meta-learning with an evolutionary strategy provides, on average, higher values of the resilience indicator. On the downside, this meta-learning method requires more iterations.
References
Moskalenko V., Kharchenko V., Moskalenko A., and Kuzikov B. Resilience and Resilient Systems of Artificial Intelligence: Taxonomy, models and methods, Algorithms, Vol. 16, No. 3, pp. 1–44, 2023. DOI:10.3390/a16030165.
Chakraborty A., Alam M., Dey V., Chattopadhyay A. and D. Mukhopadhyay, A survey on adversarial attacks and defences, CAAI Transactions on Intelligence Technology, 2021.Vol. 6, No. 1, pp. 25–45. DOI:10.1049/cit2.12028.
Hoang L.-H., Hanif M. A. and Shafique M. Tre-map: Towards reducing the overheads of fault-aware retraining of deep neural networks by merging fault maps, Proceedings of the 2021 24th Euromicro Conference on Digital System Design (DSD). Palermo, Italy, 1–3 September 2021, 8 p. DOI:10.1109/dsd53832.2021.00072.
Lu J., Liu A., Dong F., Gu F., Gama J. and Zhang G., Learning under Concept Drift: A Review, IEEE Transactions on Knowledge and Data Engineering, 2019, Vol. 31, No. 12, pp. 2346–2363, DOI: 10.1109/TKDE.2018.2876857.
Dymond J., Graceful degradation and related fields, ePrints Soton, https://eprints.soton.ac.uk/455349/ (accessed May 18, 2023).
Eigner O., Xu K., Liu S., Chen Pin-Yu, Weng Tsui-Wei, Gan Ch. and Wang M., Towards resilient artificial intelligence: Survey and research issues, 2021 IEEE International Conference on Cyber Security and Resilience (CSR), 2021. Rhodes, Greece, 26–28 July 2021. pp. 536– 542. DOI:10.1109/csr51186.2021.9527986.
Wang R., Xu K. , Liu S., Chen Pin-Yu et al. On Fast Adversarial Robustness Adaptation in Model-Agnostic Meta-Learning, ArXiv, Vol. abs/2102.10454, 2021, pp. 1– 16. DOI: 10.48550/arXiv.2102.10454.
Son X., Yang Y., Choromanski K., Caluwaerts K., Gao W., Finn C. and Tan J. Rapidly adaptable legged robots via evolutionary meta-learning, 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Las Vegas, NV, USA, 24 October 2020 – 24 January 2021, pp. 1–11. DOI:10.1109/iros45743.2020.9341571.
Ding N., Qin Y., Yang G. et al. Parameter-efficient finetuning of large-scale pre-trained language models, Nature Machine Intelligence, 2023, Vol. 5, No. 3, pp. 220–235. DOI:10.1038/s42256-023-00626-4.
Fraccascia L., Giannoccaro I., and Albino V. Resilience of Complex Systems: State of the art and directions for future research, Complexity, 2018, Vol. 2018, pp. 1–44. DOI:10.1155/2018/3421529.
Drozd O., Kharchenko V., Rucinski A., Kochanski T., Garbos R. and Maevsky D. Development of models in resilient computing, 2019 10th International Conference on Dependable Systems, Services and Technologies (DESSERT), 2019, Leeds, United Kingdom, 5–7 June 2019, pp. 1–6. DOI:10.1109/dessert.2019.8770035.
Inouye B. D., Brosi B. J., Le Sage E. H., and M. T. Lerdau, Trade-offs among resilience, robustness, stability, and performance and how we might study them, Integrative and Comparative Biology, 2021, Vol. 61, No. 6, pp. 2180–2189. DOI:10.1093/icb/icab178.
Santos S. G. T. d. C., Gonçalves Júnior P. M. , Silva G. D. d. S. and de Barros R. S. M. Speeding Up Recovery from Concept Drifts, Machine Learning and Knowledge Discovery in Databases. Berlin, Heidelberg: Springer Berlin Heidelberg, 2014, pp. 179–194. DOI: 10.1007/978-3-66244845-8_12.
Xie C., Wang J., Zhang Z., Ren Z. and Yuille A. Mitigating Adversarial Effects Through Randomization, Proceedings of the International Conference on Learning Representations, Toulon. France, 24–26 April 2017, pp. 1–16. DOI: 10.48550/arXiv.1711.01991.
Kwon H., and Lee J. Diversity Adversarial Training against Adversarial Attack on Deep Neural Networks, Symmetry, 2021, Vol. 13, No. 3, pp. 1–14, DOI: 10.3390/sym13030428.
Abusnaina A., Wu Y., Arora S. et al. Adversarial example detection using latent neighborhood graph, 2021 IEEE/CVF International Conference on Computer Vision (ICCV). Montreal, QC, Canada, 10–17 October 2021, pp. 7687– 7696. DOI:10.1109/iccv48922.2021.00759.
Li W., Ning X., Ge G., Chen X., Wang Y. and Yang H. FTT-NAS: Discovering Fault-Tolerant Neural Architecture, 2020 25th Asia South Pacific Des. Automat. Conf. (ASP-DAC). Beijing, China, 13–16 Jan. 2020. IEEE, 2020, pp. 2011–2016. DOI: 10.1109/aspdac47756.2020.9045324.
Xu H., Chen Z., Wu W., Jin Z., Kuo S.-Y. and Lyu M. NVDNN: Towards Fault-Tolerant DNN Systems with NVersion Programming, 2019 49th Annu. IEEE/IFIP Int. Conf. Dependable Syst. Netw. Workshops (DSN-W). Portland, OR, USA, 24–27 Jun. 2019. IEEE, 2019, pp. 44– 47. DOI: 10.1109/dsn-w.2019.00016.
Javaheripi M. and Koushanfar F. HASHTAG: Hash Signatures for Online Detection of Fault-Injection Attacks on Deep Neural Networks, 2021 IEEE/ACM Int. Conf. Comput. Aided Des. (ICCAD). Munich, Germany, 1–4 Nov. 2021. IEEE, 2021, pp. 1–9. DOI: 10.1109/iccad51958.2021.9643556.
Volpi R., Namkoong H., Sener O.et al. Generalizing to unseen domains via adversarial data augmentation, Proceedings of the 32nd International Conference on Neural Information Processing Systems. Montréal, QC, Canada, 2–8 December 2018, pp. 1–11. DOI: 10.5555/3327345.3327439.
Kulinchenko H. V., Drozdenko O. O., Leontiev P. V. and Hrek V. M. Pressure Regulator for Low Temperature Separation Process, 2021 IEEE 12th Int. Conf. Electron. Inf. Technol. (ELIT). Lviv, Ukraine, 19–21 May 2021. IEEE, 2021, pp. 315–319. DOI: 10.1109/elit53502.2021.9501143.
An W., Wang H. , Sun Q., Xu J., Dai Q. and Zhang L. A PID controller approach for stochastic optimization of Deep Networks, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, Salt Lake City, UT, 18–23 June 2018, pp. 8522–8531. DOI:10.1109/cvpr.2018.00889.
Tian Y., Zhao X. and Huangm W. Meta-learning approaches for learning-to-learn in deep learning: A survey, Neurocomputing, 2022, Vol. 494, pp. 203–223. DOI: 10.1016/j.neucom.2022.04.078.
Yang X. and Xu J., Few-shot Classification with First-order Task Agnostic Meta-learning, 2022 3rd International Conference on Computer Vision, Image and Deep Learning & International Conference on Computer Engineering and Applications (CVIDL & ICCEA). Changchun, China, 20–22 May 2022, pp. 2017–2020. DOI:10.1109/cvidliccea56201.2022.9824307.
Song X., Yang Y., Choromanski K. et al. Rapidly Adaptable Legged Robots via Evolutionary Meta-Learning, 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Las Vegas, NV, USA, 24 October 2020 – 24 January 2021, pp. 3769–3776. DOI:10.1109/iros45743.2020.9341571.
Wheaton M. and Madni Azad M. Resiliency and Affordability Attributes in a System Tradespace, AIAA SPACE 2015 Conference and Exposition, Pasadena, California. Reston, Virginia, 31 Aug-2 Sep 2015. DOI:10.2514/6.2015-4434.
Bansal T., Alzubi S., Wang T., Lee Jay-Yoon and McCallum A., Meta-Adapters: Parameter Efficient Few-shot Fine-tuning through Meta-Learning, First Conference on Automated Machine Learning, 2022, 18 p. Available at: https://openreview.net/pdf?id=bQt8dWKsfso.
Jiang Z., Jiang Z., Mao Ch. et al. Rethinking Efficient Tuning Methods from a Unified Perspective (Version 1), arXiv, 2023. DOI:10.48550/ARXIV.2303.00690.
Chen H., Tao R., Zhang H. et al. Conv-Adapter: Exploring Parameter Efficient Transfer Learning for ConvNets (Version 3), arXiv, 2022. DOI:10.48550/ARXIV.2208.07463.
Wu N., Hou H., Jia X., Chang X. and Li H. Low-Resource Neural Machine Translation Based on Improved Reptile Meta-learning Method, Communications in Computer and Information Science. Singapore, 2021. pp. 39–50. DOI: 10.1007/978-981-16-7512-6_4.
Park S. and So J., On the Effectiveness of Adversarial Training in Defending against Adversarial Example Attacks for Image Classification, Applied Sciences, 2020, Vol. 10, No. 22, pp. 1–16. DOI: 10.3390/app10228079.
Kotyan Sh., and Vargas D. Vasconcellos, Adversarial robustness assessment: Why in evaluation both L0 and L∞ attacks are necessary, PLOS ONE, 2022, Vol. 17, No. 4, pp. e0265723, DOI: 10.1371/journal.pone.0265723.
Li G., Pattabiraman K.and DeBardeleben N. TensorFI: A Configurable Fault Injector for TensorFlow Applications, 2018 IEEE International Symposium on Software Reliability Engineering Workshops (ISSREW). Memphis, TN, 15–18 October 2018, pp. 1–8. DOI: 10.1109/issrew.2018.00024.
Foldy-Porto T., Venkatesha Y. and Panda P. Activation Density Driven Efficient Pruning in Training, 2020 25th International Conference on Pattern Recognition (ICPR), Milan, Italy, 10–15 January 2021, pp. 8929–8936. DOI: 10.1109/icpr48806.2021.9413182.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2023 В. В. Москаленко
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
Creative Commons Licensing Notifications in the Copyright Notices
The journal allows the authors to hold the copyright without restrictions and to retain publishing rights without restrictions.
The journal allows readers to read, download, copy, distribute, print, search, or link to the full texts of its articles.
The journal allows to reuse and remixing of its content, in accordance with a Creative Commons license СС BY -SA.
Authors who publish with this journal agree to the following terms:
-
Authors retain copyright and grant the journal right of first publication with the work simultaneously licensed under a Creative Commons Attribution License CC BY-SA that allows others to share the work with an acknowledgement of the work's authorship and initial publication in this journal.
-
Authors are able to enter into separate, additional contractual arrangements for the non-exclusive distribution of the journal's published version of the work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgement of its initial publication in this journal.
-
Authors are permitted and encouraged to post their work online (e.g., in institutional repositories or on their website) prior to and during the submission process, as it can lead to productive exchanges, as well as earlier and greater citation of published work.
