EarlyExodus: Leveraging early exits to mitigate backdoor vulnerability in deep learning
The rapid migration of artificial-intelligence workloads toward edge computing significantly enhances capabilities in critical applications such as autonomous vehicles, augmented and virtual reality, and e-health, but it also heightens the urgency for robust security. However, this urgency reveals a...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-09-01
|
| Series: | Results in Engineering |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2590123025024740 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849728388180213760 |
|---|---|
| author | Salmane Douch M. Riduan Abid Khalid Zine-Dine Driss Bouzidi Fatima Ezzahra El Aidos Driss Benhaddou |
| author_facet | Salmane Douch M. Riduan Abid Khalid Zine-Dine Driss Bouzidi Fatima Ezzahra El Aidos Driss Benhaddou |
| author_sort | Salmane Douch |
| collection | DOAJ |
| description | The rapid migration of artificial-intelligence workloads toward edge computing significantly enhances capabilities in critical applications such as autonomous vehicles, augmented and virtual reality, and e-health, but it also heightens the urgency for robust security. However, this urgency reveals a critical gap: state-of-the-art backdoor defenses remain vulnerable to sophisticated data-poisoning attacks that subtly embed malicious triggers into training data and covertly manipulate model predictions, threatening the reliability and trustworthiness of edge-deployed AI.To counter this threat, we propose a defense mechanism that neutralizes advanced data poisoning attacks, clearly identifies maliciously targeted labels, and preserves model accuracy and integrity across diverse architectures and datasets. Our technique, EarlyExodus, integrates early-exit branches within neural networks and trains them with a divergence objective so that, for poisoned inputs, the early exit exposes the malicious label while the final exit maintains the correct classification. Extensive experiments on LeNet-5, ResNet-32, and GhostNet across MNIST, CIFAR-10, and GTSRB reduce the average attack success rate of seven recent backdoor attacks to about 3%, with clean-data accuracy degradation kept below 2%. These results demonstrate a practical, architecture-agnostic pathway toward trustworthy edge-AI systems and lay the foundation for extending backdoor defenses beyond image models to broader application domains. |
| format | Article |
| id | doaj-art-5d0d70b9c23a4c36b72a3ebaa04b55d0 |
| institution | DOAJ |
| issn | 2590-1230 |
| language | English |
| publishDate | 2025-09-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Results in Engineering |
| spelling | doaj-art-5d0d70b9c23a4c36b72a3ebaa04b55d02025-08-20T03:09:34ZengElsevierResults in Engineering2590-12302025-09-012710640410.1016/j.rineng.2025.106404EarlyExodus: Leveraging early exits to mitigate backdoor vulnerability in deep learningSalmane Douch0M. Riduan Abid1Khalid Zine-Dine2Driss Bouzidi3Fatima Ezzahra El Aidos4Driss Benhaddou5National School of Computer Science and Systems Analysis (ENSIAS), Mohammed V University in Rabat, Rabat, Morocco; Corresponding author.TSYS School of Computer Science, Columbus State University, Columbus, GA, USAFaculty of Sciences (FSR), Mohammed V University in Rabat, Rabat, MoroccoNational School of Computer Science and Systems Analysis (ENSIAS), Mohammed V University in Rabat, Rabat, MoroccoCollege of Natural Sciences and Mathematics, University of Houston, Houston, USACollege of Engineering, Alfaisal University, Riyadh, Saudi ArabiaThe rapid migration of artificial-intelligence workloads toward edge computing significantly enhances capabilities in critical applications such as autonomous vehicles, augmented and virtual reality, and e-health, but it also heightens the urgency for robust security. However, this urgency reveals a critical gap: state-of-the-art backdoor defenses remain vulnerable to sophisticated data-poisoning attacks that subtly embed malicious triggers into training data and covertly manipulate model predictions, threatening the reliability and trustworthiness of edge-deployed AI.To counter this threat, we propose a defense mechanism that neutralizes advanced data poisoning attacks, clearly identifies maliciously targeted labels, and preserves model accuracy and integrity across diverse architectures and datasets. Our technique, EarlyExodus, integrates early-exit branches within neural networks and trains them with a divergence objective so that, for poisoned inputs, the early exit exposes the malicious label while the final exit maintains the correct classification. Extensive experiments on LeNet-5, ResNet-32, and GhostNet across MNIST, CIFAR-10, and GTSRB reduce the average attack success rate of seven recent backdoor attacks to about 3%, with clean-data accuracy degradation kept below 2%. These results demonstrate a practical, architecture-agnostic pathway toward trustworthy edge-AI systems and lay the foundation for extending backdoor defenses beyond image models to broader application domains.http://www.sciencedirect.com/science/article/pii/S2590123025024740Data poisoning attacksBackdoor attacksEarly exits neural networksEarlyExodus |
| spellingShingle | Salmane Douch M. Riduan Abid Khalid Zine-Dine Driss Bouzidi Fatima Ezzahra El Aidos Driss Benhaddou EarlyExodus: Leveraging early exits to mitigate backdoor vulnerability in deep learning Results in Engineering Data poisoning attacks Backdoor attacks Early exits neural networks EarlyExodus |
| title | EarlyExodus: Leveraging early exits to mitigate backdoor vulnerability in deep learning |
| title_full | EarlyExodus: Leveraging early exits to mitigate backdoor vulnerability in deep learning |
| title_fullStr | EarlyExodus: Leveraging early exits to mitigate backdoor vulnerability in deep learning |
| title_full_unstemmed | EarlyExodus: Leveraging early exits to mitigate backdoor vulnerability in deep learning |
| title_short | EarlyExodus: Leveraging early exits to mitigate backdoor vulnerability in deep learning |
| title_sort | earlyexodus leveraging early exits to mitigate backdoor vulnerability in deep learning |
| topic | Data poisoning attacks Backdoor attacks Early exits neural networks EarlyExodus |
| url | http://www.sciencedirect.com/science/article/pii/S2590123025024740 |
| work_keys_str_mv | AT salmanedouch earlyexodusleveragingearlyexitstomitigatebackdoorvulnerabilityindeeplearning AT mriduanabid earlyexodusleveragingearlyexitstomitigatebackdoorvulnerabilityindeeplearning AT khalidzinedine earlyexodusleveragingearlyexitstomitigatebackdoorvulnerabilityindeeplearning AT drissbouzidi earlyexodusleveragingearlyexitstomitigatebackdoorvulnerabilityindeeplearning AT fatimaezzahraelaidos earlyexodusleveragingearlyexitstomitigatebackdoorvulnerabilityindeeplearning AT drissbenhaddou earlyexodusleveragingearlyexitstomitigatebackdoorvulnerabilityindeeplearning |