Scholarlink

<p>The problem of adversarial attacks is becoming increasingly threatening to the quality of Artificial Intelligence (AI) systems and their integrity. This paper presents a Federated Learning (FL)-based Federated adversarial attack (FAA) identification framework, which is optimized via a novel Sine Cosine Algorithm (SCA)-based Sine Chimp Optimization Algorithm (SChOA), that combines the exploration of the Sine Cosine Algorithm (SCA) and the exploitation of the Chimp Optimization Algorithm (ChOA). Quantile normalization on the pre-processing of the data reduced the influence of outliers and the Hellinger distance on feature selection increased the model accuracy by enhancing the representation of the features. The FL-SChOA presented has been demonstrated to possess accuracy of 98.50, precision of 98.48, recall of 98.59, F1 score of 98.60 and a detection time of only 10&#xa0;s with the experimentation showing that the proposed FL-SChOA is better than any other adversarial detection method. These results contribute to the idea that the creation of hybrid optimization and FL may be employed to improve the work of AI systems in the process of adversarial attacks detection.</p>

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

Federated learning-based hybrid optimization for adversarial attack detection in artificial intelligence systems

  • 期刊论文

  • 发表日期：

    2026年5月29日

查看全文
  • Saleh Alghamdi

摘要

The problem of adversarial attacks is becoming increasingly threatening to the quality of Artificial Intelligence (AI) systems and their integrity. This paper presents a Federated Learning (FL)-based Federated adversarial attack (FAA) identification framework, which is optimized via a novel Sine Cosine Algorithm (SCA)-based Sine Chimp Optimization Algorithm (SChOA), that combines the exploration of the Sine Cosine Algorithm (SCA) and the exploitation of the Chimp Optimization Algorithm (ChOA). Quantile normalization on the pre-processing of the data reduced the influence of outliers and the Hellinger distance on feature selection increased the model accuracy by enhancing the representation of the features. The FL-SChOA presented has been demonstrated to possess accuracy of 98.50, precision of 98.48, recall of 98.59, F1 score of 98.60 and a detection time of only 10 s with the experimentation showing that the proposed FL-SChOA is better than any other adversarial detection method. These results contribute to the idea that the creation of hybrid optimization and FL may be employed to improve the work of AI systems in the process of adversarial attacks detection.