Scholarlink

Face recognition systems are critical for cyber security applications, offering solutions for authentication, surveillance, and access control. Traditional systems require extensive labeled datasets to achieve high accuracy, posing challenges when data is limited. Zero-Shot Learning (ZSL) provides an efficient alternative by enabling models to recognize previously unseen classes without exhaustive data collection. This paper presents an innovative approach to face recognition using ZSL, aiming to enhance model generalization and performance with limited data, having accuracy more than 94%. A ZSL-based face recognition model, leveraging semantic attributes from LFW, is developed and evaluated to overcome the disparity between seen and unseen classes. Our comparative analysis with traditional supervised learning models, such as AlexNet and VGG16, highlights the advantages of ZSL from the perspective of precision, efficiency, and robustness. Experimental results demonstrate the model’s ability to maintain high recognition rates for unseen classes, significantly reducing dependency on large labeled datasets. This study discusses the implications of ZSL for improving the adaptability, viability, and resilience of face recognition systems in the cyber security domain, providing valuable insights for future research and development.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Zero-Shot Learning for Face Recognition in Cyber Security: Enhancing Model Generalization with Limited Data

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Abinash Kumar Pala,
  • Manish Tripathy,
  • Rashmi Ranjan Sahoo,
  • Raghunandan Swain,
  • Dinesh Kumar Dash,
  • Sandipan Mallik

摘要

Face recognition systems are critical for cyber security applications, offering solutions for authentication, surveillance, and access control. Traditional systems require extensive labeled datasets to achieve high accuracy, posing challenges when data is limited. Zero-Shot Learning (ZSL) provides an efficient alternative by enabling models to recognize previously unseen classes without exhaustive data collection. This paper presents an innovative approach to face recognition using ZSL, aiming to enhance model generalization and performance with limited data, having accuracy more than 94%. A ZSL-based face recognition model, leveraging semantic attributes from LFW, is developed and evaluated to overcome the disparity between seen and unseen classes. Our comparative analysis with traditional supervised learning models, such as AlexNet and VGG16, highlights the advantages of ZSL from the perspective of precision, efficiency, and robustness. Experimental results demonstrate the model’s ability to maintain high recognition rates for unseen classes, significantly reducing dependency on large labeled datasets. This study discusses the implications of ZSL for improving the adaptability, viability, and resilience of face recognition systems in the cyber security domain, providing valuable insights for future research and development.