Slide
Centre Interdisciplinaire
de Recherche et d’Innovation
en Cybersécurité et Société
de Recherche et d’Innovation
en Cybersécurité et Société
1.
Allaoui, M. L.; Allili, M. S.
MixLVMM: A Mixture of Lightweight Vision Mamba Model for Enhancing Skin Lesion Segmentation Across High Tone Variability Journal Article
In: IEEE Access, vol. 13, pp. 121234–121249, 2025, ISSN: 21693536 (ISSN).
Abstract | Links | BibTeX | Tags: Attention mechanism, Attention mechanisms, Computational efficiency, Critical challenges, Dermatology, Diagnosis, Image segmentation, Lesion segmentations, Lung cancer, Mixture of experts model, Mixture-of-experts model, Segmentation performance, Skin lesion, Skin lesion segmentation, Skin/lesion tone variability, Vision mamba
@article{allaoui_mixlvmm_2025,
title = {MixLVMM: A Mixture of Lightweight Vision Mamba Model for Enhancing Skin Lesion Segmentation Across High Tone Variability},
author = {M. L. Allaoui and M. S. Allili},
url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-105012036322&doi=10.1109%2FACCESS.2025.3588476&partnerID=40&md5=1cf51dcf43653e1677ad36a1360392ac},
doi = {10.1109/ACCESS.2025.3588476},
issn = {21693536 (ISSN)},
year = {2025},
date = {2025-01-01},
journal = {IEEE Access},
volume = {13},
pages = {121234–121249},
abstract = {Accurate skin lesion segmentation remains a critical challenge in automated dermatological diagnosis due to heterogeneous lesion presentations, ambiguous boundaries, imaging artifacts, and significant variability in skin and lesion tones across diverse populations. Current segmentation methods inadequately address these multifaceted complexities, particularly failing to handle extreme tone variations that can lead to diagnostic bias. To address these limitations, we present the Mixture of Lightweight Vision Mamba Model (MixLVMM), a novel expert-based framework that enhances segmentation performance across high tone variability through specialized processing. Our approach employs a Siamese network with triplet loss as a gate mechanism to categorize lesions based on tonal characteristics, routing each image to specialized Vision Mamba Model (VMM) experts optimized for specific lesion categories. Each expert utilizes a U-shaped architecture incorporating Focused Vision Mamba blocks and Adaptive Salient Region Attention modules to capture lesion-specific features while maintaining computational efficiency. Comprehensive evaluation on ISIC and PH2 datasets demonstrates that MixLVMM achieves superior segmentation performance with an average Dice coefficient of 93%, surpassing state-of-the-art methods while maintaining efficiency with only 2.5M parameters. These results establish MixLVMM as a robust solution for addressing tone-related segmentation challenges in clinical dermatology, offering both high accuracy and practical deployment feasibility for real-world applications. © 2013 IEEE.},
keywords = {Attention mechanism, Attention mechanisms, Computational efficiency, Critical challenges, Dermatology, Diagnosis, Image segmentation, Lesion segmentations, Lung cancer, Mixture of experts model, Mixture-of-experts model, Segmentation performance, Skin lesion, Skin lesion segmentation, Skin/lesion tone variability, Vision mamba},
pubstate = {published},
tppubtype = {article}
}
Accurate skin lesion segmentation remains a critical challenge in automated dermatological diagnosis due to heterogeneous lesion presentations, ambiguous boundaries, imaging artifacts, and significant variability in skin and lesion tones across diverse populations. Current segmentation methods inadequately address these multifaceted complexities, particularly failing to handle extreme tone variations that can lead to diagnostic bias. To address these limitations, we present the Mixture of Lightweight Vision Mamba Model (MixLVMM), a novel expert-based framework that enhances segmentation performance across high tone variability through specialized processing. Our approach employs a Siamese network with triplet loss as a gate mechanism to categorize lesions based on tonal characteristics, routing each image to specialized Vision Mamba Model (VMM) experts optimized for specific lesion categories. Each expert utilizes a U-shaped architecture incorporating Focused Vision Mamba blocks and Adaptive Salient Region Attention modules to capture lesion-specific features while maintaining computational efficiency. Comprehensive evaluation on ISIC and PH2 datasets demonstrates that MixLVMM achieves superior segmentation performance with an average Dice coefficient of 93%, surpassing state-of-the-art methods while maintaining efficiency with only 2.5M parameters. These results establish MixLVMM as a robust solution for addressing tone-related segmentation challenges in clinical dermatology, offering both high accuracy and practical deployment feasibility for real-world applications. © 2013 IEEE.
2.
Allaoui, M. L.; Allili, M. S.
MixLVMM: A Mixture of Lightweight Vision Mamba Model for Enhancing Skin Lesion Segmentation Across High Tone Variability Journal Article
In: IEEE Access, vol. 13, pp. 121234–121249, 2025, ISSN: 21693536 (ISSN).
Abstract | Links | BibTeX | Tags: Attention mechanism, Attention mechanisms, Computational efficiency, Critical challenges, Dermatology, Diagnosis, Image segmentation, Lesion segmentations, Lung cancer, Mixture of experts model, Mixture-of-experts model, Segmentation performance, Skin lesion, Skin lesion segmentation, Skin/lesion tone variability, Vision mamba
@article{allaouiMixLVMMMixtureLightweight2025,
title = {MixLVMM: A Mixture of Lightweight Vision Mamba Model for Enhancing Skin Lesion Segmentation Across High Tone Variability},
author = {M. L. Allaoui and M. S. Allili},
url = {https://www.scopus.com/pages/publications/105012036322?origin=resultslist},
doi = {10.1109/ACCESS.2025.3588476},
issn = {21693536 (ISSN)},
year = {2025},
date = {2025-01-01},
journal = {IEEE Access},
volume = {13},
pages = {121234–121249},
publisher = {Institute of Electrical and Electronics Engineers Inc.},
abstract = {Accurate skin lesion segmentation remains a critical challenge in automated dermatological diagnosis due to heterogeneous lesion presentations, ambiguous boundaries, imaging artifacts, and significant variability in skin and lesion tones across diverse populations. Current segmentation methods inadequately address these multifaceted complexities, particularly failing to handle extreme tone variations that can lead to diagnostic bias. To address these limitations, we present the Mixture of Lightweight Vision Mamba Model (MixLVMM), a novel expert-based framework that enhances segmentation performance across high tone variability through specialized processing. Our approach employs a Siamese network with triplet loss as a gate mechanism to categorize lesions based on tonal characteristics, routing each image to specialized Vision Mamba Model (VMM) experts optimized for specific lesion categories. Each expert utilizes a U-shaped architecture incorporating Focused Vision Mamba blocks and Adaptive Salient Region Attention modules to capture lesion-specific features while maintaining computational efficiency. Comprehensive evaluation on ISIC and PH2 datasets demonstrates that MixLVMM achieves superior segmentation performance with an average Dice coefficient of 93%, surpassing state-of-the-art methods while maintaining efficiency with only 2.5M parameters. These results establish MixLVMM as a robust solution for addressing tone-related segmentation challenges in clinical dermatology, offering both high accuracy and practical deployment feasibility for real-world applications. © 2013 IEEE.},
keywords = {Attention mechanism, Attention mechanisms, Computational efficiency, Critical challenges, Dermatology, Diagnosis, Image segmentation, Lesion segmentations, Lung cancer, Mixture of experts model, Mixture-of-experts model, Segmentation performance, Skin lesion, Skin lesion segmentation, Skin/lesion tone variability, Vision mamba},
pubstate = {published},
tppubtype = {article}
}
Accurate skin lesion segmentation remains a critical challenge in automated dermatological diagnosis due to heterogeneous lesion presentations, ambiguous boundaries, imaging artifacts, and significant variability in skin and lesion tones across diverse populations. Current segmentation methods inadequately address these multifaceted complexities, particularly failing to handle extreme tone variations that can lead to diagnostic bias. To address these limitations, we present the Mixture of Lightweight Vision Mamba Model (MixLVMM), a novel expert-based framework that enhances segmentation performance across high tone variability through specialized processing. Our approach employs a Siamese network with triplet loss as a gate mechanism to categorize lesions based on tonal characteristics, routing each image to specialized Vision Mamba Model (VMM) experts optimized for specific lesion categories. Each expert utilizes a U-shaped architecture incorporating Focused Vision Mamba blocks and Adaptive Salient Region Attention modules to capture lesion-specific features while maintaining computational efficiency. Comprehensive evaluation on ISIC and PH2 datasets demonstrates that MixLVMM achieves superior segmentation performance with an average Dice coefficient of 93%, surpassing state-of-the-art methods while maintaining efficiency with only 2.5M parameters. These results establish MixLVMM as a robust solution for addressing tone-related segmentation challenges in clinical dermatology, offering both high accuracy and practical deployment feasibility for real-world applications. © 2013 IEEE.
3.
Moudoud, H.; Houda, Z. A. El; Brik, B.; Jan, M. A.; Alshawi, B.
Advancing Robustness and Privacy in Federated Learning for Secure Autonomous Vehicle Systems Journal Article
In: IEEE Transactions on Consumer Electronics, vol. 71, no. 2, pp. 6183–6192, 2025, ISSN: 00983063 (ISSN).
Abstract | Links | BibTeX | Tags: Augmented intelligence of things, Autonomous driving, Autonomous vehicle system, Autonomous Vehicles, Critical challenges, Differential privacy, Distributed machine learning, Driving environment, Federated learning, Privacy, Security, Security and privacy, Vehicle system
@article{moudoudAdvancingRobustnessPrivacy2025,
title = {Advancing Robustness and Privacy in Federated Learning for Secure Autonomous Vehicle Systems},
author = {H. Moudoud and Z. A. El Houda and B. Brik and M. A. Jan and B. Alshawi},
url = {https://www.scopus.com/pages/publications/105002620945?origin=resultslist},
doi = {10.1109/TCE.2025.3558999},
issn = {00983063 (ISSN)},
year = {2025},
date = {2025-01-01},
journal = {IEEE Transactions on Consumer Electronics},
volume = {71},
number = {2},
pages = {6183–6192},
publisher = {Institute of Electrical and Electronics Engineers Inc.},
abstract = {The rapid development of Autonomous Vehicle Systems (AVS) is transforming transportation, enabling safer, more efficient mobility. However, ensuring the security and privacy of sensitive data generated by AVS remains a major challenge. Federated Learning (FL) has emerged as a promising solution for AVS by enabling distributed machine learning across connected vehicles without sharing raw data, thereby enhancing privacy. Despite these advantages, FL faces critical challenges in autonomous driving environments, including high communication overhead, latency, and vulnerability to adversarial attacks. To address these challenges, we propose SecureFL, a novel framework designed to enhance the robustness and privacy of FL in autonomous vehicle systems. First, we propose a Federated Gradient Sign Attack (FGSA) detection mechanism using an ensemble of classifiers to identify and mitigate adversarial attacks that attempt to corrupt the global learning model. Then, we integrate a Graph Neural Network (GNN)-based reputation system that evaluates the reliability of vehicles based on data quality, prioritizing contributions from trustworthy sources, and dynamically adjusting participation in the FL process. Finally, we introduce an uplink scheduling mechanism utilizing a rate-splitting multiple access (RSMA) technique to optimize data transmission and reduce latency, ensuring efficient communication across the AVS network. The framework’s effectiveness is validated through simulations in real-world AVS environments, demonstrating SecureFL’s capability to strengthen security, privacy, and communication efficiency in federated learning for autonomous vehicles. This work contributes to advancing the robustness and privacy of FL, enabling safer and more secure autonomous driving. © 1975-2011 IEEE.},
keywords = {Augmented intelligence of things, Autonomous driving, Autonomous vehicle system, Autonomous Vehicles, Critical challenges, Differential privacy, Distributed machine learning, Driving environment, Federated learning, Privacy, Security, Security and privacy, Vehicle system},
pubstate = {published},
tppubtype = {article}
}
The rapid development of Autonomous Vehicle Systems (AVS) is transforming transportation, enabling safer, more efficient mobility. However, ensuring the security and privacy of sensitive data generated by AVS remains a major challenge. Federated Learning (FL) has emerged as a promising solution for AVS by enabling distributed machine learning across connected vehicles without sharing raw data, thereby enhancing privacy. Despite these advantages, FL faces critical challenges in autonomous driving environments, including high communication overhead, latency, and vulnerability to adversarial attacks. To address these challenges, we propose SecureFL, a novel framework designed to enhance the robustness and privacy of FL in autonomous vehicle systems. First, we propose a Federated Gradient Sign Attack (FGSA) detection mechanism using an ensemble of classifiers to identify and mitigate adversarial attacks that attempt to corrupt the global learning model. Then, we integrate a Graph Neural Network (GNN)-based reputation system that evaluates the reliability of vehicles based on data quality, prioritizing contributions from trustworthy sources, and dynamically adjusting participation in the FL process. Finally, we introduce an uplink scheduling mechanism utilizing a rate-splitting multiple access (RSMA) technique to optimize data transmission and reduce latency, ensuring efficient communication across the AVS network. The framework’s effectiveness is validated through simulations in real-world AVS environments, demonstrating SecureFL’s capability to strengthen security, privacy, and communication efficiency in federated learning for autonomous vehicles. This work contributes to advancing the robustness and privacy of FL, enabling safer and more secure autonomous driving. © 1975-2011 IEEE.
4.
Allaoui, M. L.; Allili, M. S.
MixLVMM: A Mixture of Lightweight Vision Mamba Model for Enhancing Skin Lesion Segmentation Across High Tone Variability Journal Article
In: IEEE Access, vol. 13, pp. 121234–121249, 2025, ISSN: 21693536 (ISSN).
Abstract | Links | BibTeX | Tags: Attention mechanism, Attention mechanisms, Computational efficiency, Critical challenges, Dermatology, Diagnosis, Image segmentation, Lesion segmentations, Lung cancer, Mixture of experts model, Mixture-of-experts model, Segmentation performance, Skin lesion, Skin lesion segmentation, Skin/lesion tone variability, Vision mamba
@article{allaouiMixLVMMMixtureLightweight2025a,
title = {MixLVMM: A Mixture of Lightweight Vision Mamba Model for Enhancing Skin Lesion Segmentation Across High Tone Variability},
author = {M. L. Allaoui and M. S. Allili},
url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-105012036322&doi=10.1109%2FACCESS.2025.3588476&partnerID=40&md5=1cf51dcf43653e1677ad36a1360392ac},
doi = {10.1109/ACCESS.2025.3588476},
issn = {21693536 (ISSN)},
year = {2025},
date = {2025-01-01},
journal = {IEEE Access},
volume = {13},
pages = {121234–121249},
abstract = {Accurate skin lesion segmentation remains a critical challenge in automated dermatological diagnosis due to heterogeneous lesion presentations, ambiguous boundaries, imaging artifacts, and significant variability in skin and lesion tones across diverse populations. Current segmentation methods inadequately address these multifaceted complexities, particularly failing to handle extreme tone variations that can lead to diagnostic bias. To address these limitations, we present the Mixture of Lightweight Vision Mamba Model (MixLVMM), a novel expert-based framework that enhances segmentation performance across high tone variability through specialized processing. Our approach employs a Siamese network with triplet loss as a gate mechanism to categorize lesions based on tonal characteristics, routing each image to specialized Vision Mamba Model (VMM) experts optimized for specific lesion categories. Each expert utilizes a U-shaped architecture incorporating Focused Vision Mamba blocks and Adaptive Salient Region Attention modules to capture lesion-specific features while maintaining computational efficiency. Comprehensive evaluation on ISIC and PH2 datasets demonstrates that MixLVMM achieves superior segmentation performance with an average Dice coefficient of 93%, surpassing state-of-the-art methods while maintaining efficiency with only 2.5M parameters. These results establish MixLVMM as a robust solution for addressing tone-related segmentation challenges in clinical dermatology, offering both high accuracy and practical deployment feasibility for real-world applications. © 2013 IEEE.},
keywords = {Attention mechanism, Attention mechanisms, Computational efficiency, Critical challenges, Dermatology, Diagnosis, Image segmentation, Lesion segmentations, Lung cancer, Mixture of experts model, Mixture-of-experts model, Segmentation performance, Skin lesion, Skin lesion segmentation, Skin/lesion tone variability, Vision mamba},
pubstate = {published},
tppubtype = {article}
}
Accurate skin lesion segmentation remains a critical challenge in automated dermatological diagnosis due to heterogeneous lesion presentations, ambiguous boundaries, imaging artifacts, and significant variability in skin and lesion tones across diverse populations. Current segmentation methods inadequately address these multifaceted complexities, particularly failing to handle extreme tone variations that can lead to diagnostic bias. To address these limitations, we present the Mixture of Lightweight Vision Mamba Model (MixLVMM), a novel expert-based framework that enhances segmentation performance across high tone variability through specialized processing. Our approach employs a Siamese network with triplet loss as a gate mechanism to categorize lesions based on tonal characteristics, routing each image to specialized Vision Mamba Model (VMM) experts optimized for specific lesion categories. Each expert utilizes a U-shaped architecture incorporating Focused Vision Mamba blocks and Adaptive Salient Region Attention modules to capture lesion-specific features while maintaining computational efficiency. Comprehensive evaluation on ISIC and PH2 datasets demonstrates that MixLVMM achieves superior segmentation performance with an average Dice coefficient of 93%, surpassing state-of-the-art methods while maintaining efficiency with only 2.5M parameters. These results establish MixLVMM as a robust solution for addressing tone-related segmentation challenges in clinical dermatology, offering both high accuracy and practical deployment feasibility for real-world applications. © 2013 IEEE.