<?xml version="1.0" encoding="UTF-8"?>
<ArticleSet>
  <Article>
    <Journal>
      <PublisherName></PublisherName>
      <JournalTitle>Journal of Artificial Intelligence, Applications and Innovations</JournalTitle>
      <Issn>3060-7124</Issn>
      <Volume></Volume>
      <Issue>In Press</Issue>
      <PubDate PubStatus="epublish">
        <Year>2026</Year>
        <Month>11</Month>
        <Day>19</Day>
      </PubDate>
    </Journal>
    <ArticleTitle>Learning with Fewer Labels, Without Sharing Raw Data: A Lightweight Self-Supervised Federated Transformer for IoT Intrusion Detection at the Edge</ArticleTitle>
    <VernacularTitle>Learning with Fewer Labels, Without Sharing Raw Data: A Lightweight Self-Supervised Federated Transformer for IoT Intrusion Detection at the Edge</VernacularTitle>
    <FirstPage></FirstPage>
    <LastPage></LastPage>
    <Language>EN</Language>
    <AuthorList>
      <Author>
        <FirstName></FirstName>
        <LastName></LastName>
        <Affiliation></Affiliation>
      </Author>
      <Author>
        <FirstName></FirstName>
        <LastName></LastName>
        <Affiliation></Affiliation>
      </Author>
      <Author>
        <FirstName></FirstName>
        <LastName></LastName>
        <Affiliation></Affiliation>
      </Author>
      <Author>
        <FirstName></FirstName>
        <LastName></LastName>
        <Affiliation></Affiliation>
      </Author>
    </AuthorList>
    <PublicationType>Journal Article</PublicationType>
    <Abstract>&lt;p&gt;Internet of Things (IoT) ecosystems have grown quickly and dramatically, creating an enormous cybersecurity attack surface that exposes highly distributed networks to sophisticated attacks. However, traditional centralized intrusion detection systems (IDS) and conventional transformer-based models have significant drawbacks in edge environments, such as high computational complexity, privacy issues, and reliance on large amounts of labeled data. To tackle these problems, this study proposes a lightweight and self-supervised federated transformer framework for edge-based intrusion detection in IoT networks. The proposed method combines a multi-task self-supervised pre-training approach to learn discriminative features from unlabeled traffic data to alleviate the lack of labeled data. Given the limited resources of edge devices, an attention-head pruning and INT8 quantization-based compression-aware optimization mechanism is proposed. Moreover, a federated learning paradigm with an effective aggregation protocol is designed to allow for distributed training without sharing raw network traffic. The framework incorporates a robust aggregation architecture designed to handle statistical heterogeneity and establish a structural foundation for resilience against unreliable client updates. The comprehensive evaluations show that the proposed framework achieves an overall classification accuracy of 99.90% and an F1-score of 0.9990 in a centralized setting, and a highly competitive accuracy of 99.87% and an F1-score of 0.9987 under distributed federated conditions. The compression strategy successfully reduces the memory usage of the model by around 79.7% and provides an inference latency of 0.0465ms per sample, demonstrating the computational efficiency and applicability of the model to real-time edge deployment.&lt;/p&gt;</Abstract>
    <ObjectList>
      <Object Type="keyword">
        <Param Name="value">Internet of Things (IoT)</Param>
      </Object>
      <Object Type="keyword">
        <Param Name="value">Intrusion Detection System (IDS)</Param>
      </Object>
      <Object Type="keyword">
        <Param Name="value">Lightweight Transformer</Param>
      </Object>
      <Object Type="keyword">
        <Param Name="value">Self-Supervised Learning</Param>
      </Object>
      <Object Type="keyword">
        <Param Name="value">Federated Learning</Param>
      </Object>
      <Object Type="keyword">
        <Param Name="value">Model Compression</Param>
      </Object>
    </ObjectList>
    <ArchiveCopySource DocType="pdf"></ArchiveCopySource>
  </Article>
</ArticleSet>
