D.Yu. Romanyuta1, A. V. Kovalenko2, A. V. Ovsyannikova3

1,2 Kuban State University (Krasnodar, Russia)
3 Financial University under the Government of the Russian Federation (Moscow, Russia)
1 dm.romanyuta@yandex.ru, 2 savanna-05@mail.ru, 3anna_ovsyannikov@bk.ru

When developing neural network models for face recognition, a pressing problem is related to the learning rate. It is necessary that the model can be trained using parallel methods on any architecture.

Increase the efficiency of developing neural network models for face recognition using existing parallel learning methods.

Various approaches to parallel implementation of Arcface model training are considered, including distributed training, multi-threaded computing and the use of GPUs. An analysis of the effectiveness of various approaches is presented and the Arcface model is trained using parallel methods.

The considered method of parallel training of a neural network model should speed up the learning process, which makes it possible to implement more complex architectures, for example, the Arcface architecture for face recognition.

Romanyuta D.Yu., Kovalenko A.V., Ovsyannikova A.V. Implementing Arcface model training using parallel methods. Nonlinear World. 2024. V. 22. № 4. P. 80–85. DOI: https://doi.org/10.18127/ j20700970-202404-10 (In Russian)

1. Kabyshev O.A., Maslakov M.P., Kabyshev A.M. Programmnaya realizaciya algoritma obucheniya nejronnoj seti. Inzhenernyj vestnik Dona. 2021. № 3. URL: ivdon.ru/ru/magazine/archive/n3y2021/6850/ (In Russian).
2. Arum R.K., Solayappan V. A., Sree S.S. Masked Deep Face Recognition using ArcFace and Ensemble Learning. Conference on Biomedical Engineering and Sciences. IEEE. 2021. URL: ieeexplore.ieee.org/document/9768777, Date accessed 20.05.2024.
3. ArcFace Knows the Gender, Too!. URL: arxiv.org/abs/2112.10101, Date accessed 22.05.2024.
4. Diffusion-Convolutional Neural Networks. URL: arxiv.org/abs/1511.02136, Date accessed 23.05.2024.
5. ArcFace: Additive Angular Margin Loss for Deep Face Recognition. URL: arxiv.org/abs/1801.07698, Date accessed 24.05.2024.
6. CosFace: Large Margin Cosine Loss for Deep Face Recognition. URL: arxiv.org/abs/1801.09414, Date accessed 24.05.2024.
7. SphereFace: Deep Hypersphere Embedding for Face Recognition. URL: arxiv.org/abs/1704.08063, Date accessed 25.05.2024.
8. A novel method for iris recognition using BP neural network and parallel computing by the aid of GPUs. URL: arxiv.org/pdf/2309.033901, Date accessed 26.05.2024.
9. Face Recognition-Based Mass Attendance Using YOLOv5 and ArcFace. URL: researchgate.net/publication/371480244_Face_ Recognition-Based_Mass_Attendance _Using_YOLOv5_and_ArcFace, Date accessed 27.05.2024.
10. Smart home Management System with Face Recognition Based on ArcFace Model in Deep Convolutional Neural Network, URL: researchgate.net/publication/365362063_Smart_home_Management_System_with_Face_Recognition_based_on_ArcFace_model_ in_Deep_Convolutional_Neural_Network, Date accessed 28.05.2024.
11. Romanyuta D.Yu., Lukashchik E.P. Nejrosetevaya sistema biometricheskoj identifikacii lichnosti po setchatke glaza. Prikladnaya matematika: sovremennye problemy matematiki, informatiki i modelirovaniya. Krasnodar: FGBU «Rossijskoe energeticheskoe agentstvo» Minenergo Rossii Krasnodarskij CNTI- filial FGBU «REA» Minenergo Rossii. 2022 (In Russian).
12. Kovalenko A.V. Nejronnaya set' i nechetkie mnozhestva, kak instrument ocenki kreditosposobnosti zaemshchika. Materialy VI ob"edinennoj nauchnoj konferencii studentov i aspirantov fakul'teta prikladnoj matematiki. Krasnodar. 2006. S. 56–58 (In Russian).
13. Romanyuta D.Yu., Kovalenko A.V., Kalajdina G.V. Sravnenie diffuzionnyh svertochnyh setej v reshenie zadachi raspoznavaniya lic. Prikladnaya matematika: sovremennye problemy matematiki, informatiki i modelirovaniya. Krasnodar. 2024 (In Russian).