Computational nanotechnologyComputational nanotechnologyComputational nanotechnology2313-223X2587-9693YUR-VAK30971510.33693/2313-223X-2025-12-2-28-36QGDHYMCYBERSECURITYКИБЕРБЕЗОПАСНОСТЬResearch ArticleAnalysis of the effectiveness and robustness of neural networks with early exits in computer vision tasksАнализ эффективности и робастности нейронных сетей с ранними выходами в задачах компьютерного зренияhttps://orcid.org/0000-0002-1154-61514334-5520ChesalinAlexander N.ЧесалинАлександр Николаевич
Russian Federation

Cand. Sci. (Eng.), Associate Professor; Head, Department of Computer and Information Security, Institute of Artificial Intelligence

кандидат технических наук, доцент; заведующий, кафедра компьютерной и информационной безопасности, Институт искусственного интеллекта

chesalin_an@mail.ru4948-2180StavtsevAlexey V.СтавцевАлексей Вячеславович
Russian Federation

Cand. Sci. (Phys.-Math.); associate professor, Department of Computer and Information Security, Institute of Artificial Intelligence

кандидат физико-математических наук; доцент, кафедра компьютерной и информационной безопасности, Институт искусственного интеллекта

stavcev@mirea.ru1935-5513UshkovaNadezhda N.УшковаНадежда Николаевна
Russian Federation

senior lecturer, Department of Computer and Information Security, Institute of Artificial Intelligence

старший преподаватель, кафедра компьютерной и информационной безопасности, Институт искусственного интеллекта

ushkova@mirea.ruhttps://orcid.org/0009-0001-1450-07147264-9403CharuginValentin V.ЧаругинВалентин Валерьевич
Russian Federation

lecturer, Department of Computer and Information Security, Institute of Artificial Intelligence

преподаватель, кафедра компьютерной и информационной безопасности, Институт искусственного интеллекта

сharugin_v@mirea.ruhttps://orcid.org/0009-0003-4950-77264080-4997CharuginValery V.ЧаругинВалерий Валерьевич
Russian Federation

lecturer, Department of Computer and Information Security, Institute of Artificial Intelligence

преподаватель, кафедра компьютерной и информационной безопасности, Институт искусственного интеллекта

сharugin@mirea.ruMIREA – Russian Technological UniversityМИРЭА – Российский технологический университет19082025122283618092025Copyright ©; 2025, Yur-VAKCopyright ©; 2025, Юр-ВАК2025Yur-VAKЮр-ВАКhttps://www.urvak.ru/contacts/https://journals.rcsi.science/2313-223X/article/view/309715

Many embedded systems and Internet of Things (IoT) devices use neural network algorithms for various information processing tasks. At the same time, developers face the problem of insufficient computing resources for effective functioning, especially in real-time (pseudo-) tasks. In this regard, the urgent task is to find a balance between the quality of the results and computational complexity. One of the ways to increase the computational efficiency of neural networks is to use neural network architectures with early exits (for example, BranchyNet), which allow making decisions before passing through all layers of the neural network, depending on the source data for a given reliability of the results. The purpose of the study: to analyze the applicability, effectiveness and robustness of neural networks with early exits (BranchyResNet18) in computer vision tasks. The analysis is based on the GTSRB road sign dataset. The research methodology is an experimental efficiency analysis based on the calculation of the number of floating-point operations (FLOP) to obtain results with a given accuracy, and an experimental robustness analysis based on the generation of various noise effects and adversarial attacks. Research results: estimates of the effectiveness of neural networks with early exit and their robustness to unintended and intentional disturbances have been obtained.

Во многих встраиваемых системах и устройствах интернета вещей (IoT) применяются нейросетевые алгоритмы для различных задач обработки информации. При этом разработчики сталкиваются с проблемой недостаточности вычислительных ресурсов для эффективного функционирования, особенно в задачах реального (псевдо) времени. В связи с этим актуальной является задача нахождение баланса между качеством результатов и вычислительной сложностью. Одним из способов повышения вычислительной эффективности нейронных сетей, является применение архитектур нейронных сетей с ранними выходами (например, BranchyNet), позволяющие принимать решения до прохождения всех слоев нейронной сети, в зависимости от исходных данных при заданной достоверности результатов. Цель исследования: провести анализ применимости, эффективности и робастности нейронных сетей с ранними выходами (BranchyResNet18) в задачах компьютерного зрения. Анализ проводится на основе набора данных дорожных знаков GTSRB. Методология исследования представляет собой экспериментальный анализ эффективности на основе расчета количества операций с плавающей запятой (FLOP) для получения результатов с заданной точностью, и экспериментальный анализ робастности на основе генерации различных шумовых воздействий и состязательных атак. Результаты исследования: получены оценки эффективности нейронных сетей с ранним выходом и их робастность к непреднамеренным и преднамеренным возмущениям.

neural networks with early exitimproving the efficiency of neural networksrobustnessadversarial attacksнейронные сети с ранним выходомповышение эффективности нейронных сетейробастностьсостязательные атакиJacob B., Kligys S., Chen B. et al. Quantization and training of neural networks for efficient integer-arithmetic-only inference. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2018. Pp. 2704–2713. URL: https://arxiv.org/abs/1712.05877Jacob B., Kligys S., Chen B. et al. Quantization and training of neural networks for efficient integer-arithmetic-only inference // Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2018. Pp. 2704–2713. URL: https://arxiv.org/abs/1712.05877Torkunova Yu.V., Milovanov D.V. Optimization of neural networks: Methods and their comparison using the example of text mining. International Journal of Advanced Studies. 2023. Vol. 13. No. 4. Pp. 142–158. (In Rus.). DOI: 10.12731/2227-930X2023-13-4-142-158.Торкунова Ю.В., Милованов Д.В. Оптимизация нейронных сетей: методы и их сравнение на примере интеллектуального анализа текста // International Journal of Advanced Studies. 2023. Т. 13. № 4. C. 142–158. DOI: 10.12731/2227-930X2023-13-4-142-158.Han S., Pool J., Tran J., Dally W.J. Learning both weights and connections for efficient neural networks. In: Advances in Neural Information Processing Systems (NeurIPS). 2015. URL: https://arxiv.org/abs/1506.02626Han S., Pool J., Tran J., Dally W.J. Learning both weights and connections for efficient neural networks // Advances in Neural Information Processing Systems (NeurIPS). 2015. URL: https://arxiv.org/abs/1506.02626Denton E.L., Zaremba W., Bruna J. et al. Exploiting linear structure within convolutional networks for efficient evaluation. In: Advances in Neural Information Processing Systems (NeurIPS). 2014. URL: https://arxiv.org/abs/1404.0736Denton E.L., Zaremba W., Bruna J. et al. Exploiting linear structure within convolutional networks for efficient evaluation // Advances in Neural Information Processing Systems (NeurIPS). 2014. URL: https://arxiv.org/abs/1404.0736Hinton G., Vinyals O., Dean J. Distilling the knowledge in a neural network. arXiv Preprint arXiv:1503.02531. 2015. URL: https://arxiv.org/abs/1503.02531Hinton G., Vinyals O., Dean J. Distilling the knowledge in a neural network // arXiv Preprint arXiv:1503.02531. 2015. URL: https://arxiv.org/abs/1503.02531Ivanov E.A., Mamonova T.E. Comparison of neural network model compression methods when used on a microcontroller. In: Youth and modern information technologies. Proceedings of the XXI International Scientific and Practical Conference of Students, Postgraduates and Young Scientists (Tomsk, April 15–18, 2024). Tomsk, 2024. Pp. 176–180.Иванов Е.А., Мамонова Т.Е. Сравнение методов сжатия модели нейронной сети при использовании на микроконтроллере // Молодежь и современные информационные технологии: сборник трудов XXI Междунар. науч.-практ. конф. студентов, аспирантов и молодых ученых (Томск, 15–18 апреля 2024 г.). Томск, 2024. C. 176–180.Ullrich K., Meeds E., Welling M. Soft weight-sharing for neural network compression. In: International Conference on Learning Representations (ICLR). 2017. URL: https://arxiv.org/abs/1702.04008Ullrich K., Meeds E., Welling M. Soft weight-sharing for neural network compression // International Conference on Learning Representations (ICLR). 2017. URL: https://arxiv.org/abs/1702.04008Scardapane S., Scarpiniti M., Baccarelli E., Uncini A. Why should we add early exits to neural networks? arXiv Preprint arXiv:2004.12814. 2020. URL: https://arxiv.org/abs/2004.12814Scardapane S., Scarpiniti M., Baccarelli E., Uncini A. Why should we add early exits to neural networks? // arXiv Preprint arXiv:2004.12814. 2020. URL: https://arxiv.org/abs/2004.12814Cheng Y., Wang D., Zhou P., Zhang T. A Survey of model compression and acceleration for deep neural networks. arXiv Preprint arXiv:1710.09282. 2017. URL: https://arxiv.org/abs/1710.09282Cheng Y., Wang D., Zhou P., Zhang T. A Survey of model compression and acceleration for deep neural networks // arXiv Preprint arXiv:1710.09282. 2017. URL: https://arxiv.org/abs/1710.09282Bajpai D.J., Hanawal M.K. A survey of early exit deep neural networks in NLP. arXiv Preprint arXiv:2501.07670. 2025. URL: https://arxiv.org/abs/2501.07670Bajpai D.J., Hanawal M.K. A survey of early exit deep neural networks in NLP // arXiv Preprint arXiv:2501.07670. 2025. URL: https://arxiv.org/abs/2501.07670Panda P., Sengupta A., Roy K. Conditional deep learning for energy-efficient and enhanced pattern recognition. In: Design, Automation & Test in Europe Conference (DATE). 2016. Pp. 475–480.Panda P., Sengupta A., Roy K. Conditional deep learning for energy-efficient and enhanced pattern recognition // Design, Automation & Test in Europe Conference (DATE). 2016. Pp. 475–480.Teerapittayanon S., McDanel B., Kung H.T. BranchyNet: Fast inference via early exiting from deep neural networks. arXiv Preprint arXiv:1709.01686. 2017. URL: https://arxiv.org/abs/1709.01686Teerapittayanon S., McDanel B., Kung H.T. BranchyNet: Fast inference via early exiting from deep neural networks // arXiv Preprint arXiv:1709.01686. 2017. URL: https://arxiv.org/abs/1709.01686Kaya Y., Hong S., Dumitras T. Shallow-deep networks: Understanding and mitigating network overthinking. In: Proceedings of the 36th International Conference on Machine Learning (ICML). 2019. Pp. 3301–3310.Kaya Y., Hong S., Dumitras T. Shallow-deep networks: Understanding and mitigating network overthinking // Proceedings of the 36th International Conference on Machine Learning (ICML). 2019. Pp. 3301–3310.Laskaridis S., Venieris S.I., Almeida M. et al. SPINN: Synergistic progressive inference of neural networks over device and cloud. In: Proceedings of the 26th Annual International Conference on Mobile Computing and Networking (MobiCom’20). 2020. Pp. 1–15.Laskaridis S., Venieris S.I., Almeida M. et al. SPINN: Synergistic progressive inference of neural networks over device and cloud // Proceedings of the 26th Annual International Conference on Mobile Computing and Networking (MobiCom’20). 2020. Pp. 1–15.Huang G., Chen D., Li T. et al. Multi-scale dense networks for resource efficient image classification. In: International Conference on Learning Representations (ICLR). 2018. 14 p.Huang G., Chen D., Li T. et al. Multi-scale dense networks for resource efficient image classification // International Conference on Learning Representations (ICLR). 2018. 14 p.Kaya Y., Dumitras T. When does dynamic computation help early-exiting neural networks? In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). 2021. Pp. 2709–2718.Kaya Y., Dumitras T. When does dynamic computation help early-exiting neural networks? // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). 2021. Pp. 2709–2718.Venieris S.I., Laskaridis S., Lane N.D. Dynamic neural networks: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI). 2023. Vol. 45. No. 2. Pp. 2076–2098.Venieris S.I., Laskaridis S., Lane N.D. Dynamic neural networks: A survey // IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI). 2023. Vol. 45. No. 2. Pp. 2076–2098.Viola P., Jones M. Rapid object detection using a boosted cascade of simple features. CVPR. 2001. Vol. 1.Viola P., Jones M. Rapid object detection using a boosted cascade of simple features // CVPR. 2001. Vol. 1.Chesalin A.N. Application of cascade classification algorithms for improving intrusion detection systems. Nonlinear World. 2022. Vol. 20. No. 1. Pp. 24–41. (In Rus.)Чесалин А.Н. Применение каскадных алгоритмов классификации для совершенствования систем обнаружения вторжений // Нелинейный мир. 2022. Т. 20. № 1. C. 24–41.He K., Zhang X., Ren S., Sun J. Deep residual learning for image recognition. arXiv:1512.03385v1 [cs.CV]. 2015.He K., Zhang X., Ren S., Sun J. Deep residual learning for image recognition // arXiv:1512.03385v1 [cs.CV]. 2015.Sai Abhishek, Allena Venkata. Resnet18 model with sequential layer for computing accuracy on image classification dataset. International Journal of Scientific Research in Computer Science, Engineering and Information Technology. 2022. Vol. 10. Pp. 2320–2882.Sai Abhishek, Allena Venkata. Resnet18 model with sequential layer for computing accuracy on image classification dataset // International Journal of Scientific Research in Computer Science, Engineering and Information Technology. 2022. Vol. 10. Pp. 2320–2882.Stallkamp J., Schlipsing M., Salmen J., Igel C. The German traffic sign recognition benchmark: A multi-class classification competition. In: IEEE International Joint Conference on Neural Networks (IJCNN). 2011.Stallkamp J., Schlipsing M., Salmen J., Igel C. The German Traffic Sign Recognition Benchmark: A multi-class classification competition // IEEE International Joint Conference on Neural Networks (IJCNN). 2011.Sovrasov V. Ptflops: A flops counting tool for neural networks in Pytorch framework. 2024. URL: https://github.com/sovrasov/flops-counter.pytorchGonzalez R., Woods R. Digital image processing. Moscow: Technosphere, 2005, 1072 p.Гонсалес Р., Вудс Р. Цифровая обработка изображений. М.: Техносфера, 2005. 1072 с.Kim H. Torchattacks: A PyTorch repository for adversarial attacks. arXiv:2010.01950. 2020.Kim H. Torchattacks: A PyTorch repository for adversarial attacks // arXiv:2010.01950. 2020.Moosavi-Dezfooli S.-M., Fawzi A., Frossard P. DeepFool: A simple and accurate method to fool deep neural networks. arXiv:1511.04599. 2015.Moosavi-Dezfooli S.-M., Fawzi A., Frossard P. DeepFool: A simple and accurate method to fool deep neural networks // arXiv:1511.04599. 2015.Rauber J., Brendel W., Bethge M. Foolbox: A Python toolbox to benchmark the robustness of machine learning models. arXiv:1707.04131. 2017.Rauber J., Brendel W., Bethge M. Foolbox: A Python toolbox to benchmark the robustness of machine learning models // arXiv:1707.04131. 2017.Goodfellow I., Shlens J., Szegedy C. Explaining and harnessing adversarial examples. arXiv:1412.6572. 2015.Goodfellow I., Shlens J., Szegedy C. Explaining and harnessing adversarial examples // arXiv:1412.6572. 2015.