报告详情
  1. 活动首页
  2. 报告清单
  3. 报告详情
Hybrid CNN-BiLSTM Architecture with Self-Attention for Adaptive Anomaly Detection in Resource-Constrained TinyML-Enabled IoT Systems
编号：118 访问权限：仅限参会人 更新：2026-11-10 16:49:09 浏览：9次 Online

报告开始：2026年11月21日 10:45（Asia/Kolkata）

报告时间：15min

所在会场：[S4] Artificial Intelligence and Intelligent Computing [S4-7] Artificial Intelligence and Intelligent Computing

演示文件

提示：该报告下的文件权限为仅限参会人，您尚未登录，暂时无法查看。

摘要
Internet of Things (IoT) devices are being rapidly deployed at a huge scale in industrial, agricultural and healthcare ecosystems producing an unprecedented volume of heterogeneous time-series data, thus making real-time anomaly detection highly challenging under limited memory and computation constraints. We present a new Hybrid Convolutional Neural Network–Bidirectional Long Short-Term Memory (CNN-BiLSTM) architecture with multi-head self-attention and robust performance targeted for deployment on TinyML edge microcontrollers. It uses 1-D convolutional layers to capture local spatial patterns and bidirectional LSTM units to exploit long-range temporal dependencies, whilst an attention module prunes the most discriminative temporal segments. We propose a post-training INT8 quantization pipeline that reduces the model footprint by a factor of 4 with maintained classification performance to enable on-device inference with <256 KB memory budget. Experimental evaluation on three open-source IoT sensor datasets including SKAB, SMD, and NAB shows that the proposed framework achieves a classification accuracy of 96.7%, F1-score of 96.0%, recall of 96.3% which outperform state-of-the-art baselines include standalone LSTM, SVM, Random Forest, and Logistic Regression. The results demonstrate that the proposed architecture provides a reasonable trade-off between prediction accuracy and efficiency which can be utilized as a practical framework for intelligent, energy-efficient edge AI applications in domains such as smart agriculture, predictive maintenance, and remote patient monitoring.
 
关键词
TinyML; Anomaly Detection; CNN-BiLSTM; Self-Attention; IoT Edge Computing; Time-Series Classification; INT8 Quantization; Predictive Maintenance; Deep Learning.
报告人
Mohamed Faizal M K
ASSISTANT PROFESSOR K.Ramakrishnan College of Engineering；TRICHY-TAMILNADU

稿件作者
Mridul Dixit GLA University；Department of Computer Engineering and Applications
PARWATI KUMAWAT madhav university abu road rajasthan
Sivanand A Saveetha School of Engineering,Saveetha Institute of Medical and Technical Sciences,Chennai, India
NAVIN KUMAR B Saveetha School of Engineering, Saveetha Institute of Medical and Technical Sciences (SIMATS),Chennai-602105, Tamil Nadu, India
Anandakumar Haldorai Sri Eshwar College of Engineering
Dr. SivaramKrishnan M School of Engineering and Emerging Technologies,Rathinam Global Deemed to be University,Coimbatore, India
Mohamed Faizal M K K.Ramakrishnan College of Engineering；TRICHY-TAMILNADU
发表评论
验证码 看不清楚，更换一张
全部评论
重要日期

  • 会议日期

    11月20日

    2026

    至

    11月22日

    2026

  • 10月31日 2026

    初稿截稿日期

  • 11月07日 2026

    提前注册日期

  • 11月22日 2026

    注册截止日期

主办单位
United Societies of Science
承办单位
Pondicherry University
协办单位
IEEE Section
IEEE Madras Section
联系方式
登录查看完整联系方式
移动端
在手机上打开
小程序
打开微信小程序
客服
扫码或点此咨询