Please use this identifier to cite or link to this item:
https://essuir.sumdu.edu.ua/handle/123456789/90070
Or use following links to share this resource in social networks:
Tweet
Recommend this item
Title | Model and Training Method of the Resilient Image Classifier Considering Faults, Concept Drift, and Adversarial Attacks |
Authors |
Moskalenko, Viacheslav Vasylovych
Kharchenko, V. Moskalenko, Alona Serhiivna Petrov, Serhii Oleksandrovych |
ORCID |
http://orcid.org/0000-0001-6275-9803 http://orcid.org/0000-0003-3443-3990 |
Keywords |
image classification robustness resilience resilience graceful degradation adversarial attacks faults injection concept drift convolutional neural network self-learning self-knowledge distillation prototypical classifier contrastive-center loss |
Type | Article |
Date of Issue | 2022 |
URI | https://essuir.sumdu.edu.ua/handle/123456789/90070 |
Publisher | MDPI |
License | Creative Commons Attribution 4.0 International License |
Citation | Moskalenko, V.; Kharchenko, V.; Moskalenko, A.; Petrov, S. Model and Training Method of the Resilient Image Classifier Considering Faults, Concept Drift, and Adversarial Attacks. Algorithms 2022, 15, 384. https://doi.org/10.3390/a15100384 |
Abstract |
Modern trainable image recognition models are vulnerable to different types of perturbations; hence, the development of resilient intelligent algorithms for safety-critical applications remains a relevant concern to reduce the impact of perturbation on model performance. This paper proposes a model and training method for a resilient image classifier capable of efficiently functioning despite various faults, adversarial attacks, and concept drifts. The proposed model has a multi-section structure with a hierarchy of optimized class prototypes and hyperspherical class boundaries, which provides adaptive computation, perturbation absorption, and graceful degradation. The proposed training method entails the application of a complex loss function assembled from its constituent parts in a particular way depending on the result of perturbation detection and the presence of new labeled and unlabeled data. The training method implements principles of self-knowledge distillation, the compactness maximization of class distribution and the interclass gap, the compression of feature representations, and consistency regularization. Consistency regularization makes it possible to utilize both labeled and unlabeled data to obtain a robust model and implement continuous adaptation. Experiments are performed on the publicly available CIFAR-10 and CIFAR-100 datasets using model backbones based on modules ResBlocks from the ResNet50 architecture and Swin transformer blocks. It is experimentally proven that the proposed prototype-based classifier head is characterized by a higher level of robustness and adaptability in comparison with the dense layer-based classifier head. It is also shown that multi-section structure and self-knowledge distillation feature conserve resources when processing simple samples under normal conditions and increase computational costs to improve the reliability of decisions when exposed to perturbations. |
Appears in Collections: |
Наукові видання (ЕлІТ) |
Views
Argentina
1
Austria
126
Czechia
14
Ireland
533
Russia
127
Singapore
440002
Switzerland
1
Ukraine
5828
United Kingdom
2914
United States
397734
Unknown Country
880001
Downloads
Algeria
1
China
1982
Czechia
13
Finland
1
Germany
32721
Ireland
1
Lithuania
1
Singapore
1
Switzerland
1
Taiwan
1057
Tunisia
1
Ukraine
270938
United Kingdom
1
United States
440001
Unknown Country
32721
Files
File | Size | Format | Downloads |
---|---|---|---|
Moskalenko_et al._Classifier_Considering_Faults.pdf | 1.7 MB | Adobe PDF | 779441 |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.