Scholarlink

This paper introduces a novel fault detection algorithm that leverages convolutional code-based word embeddings and a seq2seq neural network architecture enhanced with a Maximum Information Coefficient (MIC)-based attention mechanism. Our approach was rigorously tested on public log datasets and demonstrated superior fault detection accuracy compared to baseline models. Key innovations include the integration of convolutional encoding to capture sequential and categorical information and the dynamic adjustment of attention weights using MIC. Ablation studies confirmed the effectiveness of these components, with the full model achieving the highest F1-Scores. This work provides a robust solution for fault detection in networked systems.

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

Deep Learning-Based Fault Detection Method Using Convolutional Code Word Embedding and Maximum Mutual Information Attention Mechanism

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yanxi Xie,
  • Xiaojun Jing,
  • Junsheng Mu

摘要

This paper introduces a novel fault detection algorithm that leverages convolutional code-based word embeddings and a seq2seq neural network architecture enhanced with a Maximum Information Coefficient (MIC)-based attention mechanism. Our approach was rigorously tested on public log datasets and demonstrated superior fault detection accuracy compared to baseline models. Key innovations include the integration of convolutional encoding to capture sequential and categorical information and the dynamic adjustment of attention weights using MIC. Ablation studies confirmed the effectiveness of these components, with the full model achieving the highest F1-Scores. This work provides a robust solution for fault detection in networked systems.