报告详情
Federated Learning with Attention-Based Bidirectional LSTM for Privacy-Preserving IoT Data Analytics in Societal Applications
编号：111
访问权限：仅限参会人
更新：2026-11-10 16:47:49 浏览：7次
Online
报告开始：2026年11月21日 14:45（Asia/Kolkata）
报告时间：15min
所在会场：[S2] Internet of Things and Intelligent Sensing Systems [S2-2] Internet of Things and Intelligent Sensing Systems
摘要
The recent expansion of Internet of Things (IoT) devices in the areas among agriculture, healthcare and industrial applications has generated a great number of heterogeneous time-series data which leads to new opportunities for intelligent decision support. Nonetheless, this approach raises serious privacy, security and regulatory issues when centralising such sensitive data for the purpose of model training. In this paper, we propose FL−AttBiLSTM-a Federated Learning (FL) framework, which utilizes an Attention based Bi-directional Long Short Term Memory (Att Bi-LSTM) network for performing IoT data analytics in privacy-aware settings. With an attention mechanism integrated, the proposed model can dynamically weight temporal dependencies learnt using the bidirectional recurrent layers to retrieve feature learning from heterogeneous distributed non-IID time-series streams across edge nodes. We show that by using a secure aggregation protocol with differential privacy (DP) noise injection at the global server, formal privacy guarantees are maintained. COMMIT: Results for (4) multi-domain IoT datasets covering the scenarios of crop health monitoring, ECG-based patient surveillance, industrial fault detection and smart city traffic analytics show that FL-AttBiLSTM achieves a mean classification accuracy of 96.7%, out-performing state-of-the-art federated and centralized baselines (3.6%−18.4% relative improvement on the average classification accuracy), while being up to +50x days more efficient (+43.75% in terms of total data transmission cost) than canonical FedAvg, thus providing new benchmarks towards practical design and deployment of reliable, scalable AI-driven IoT analytics platforms empowered by Federated Learning approaches.
关键词
Federated Learning, Bidirectional LSTM, Attention Mechanism, Internet of Things, Differential Privacy, Time-Series Classification, Data Science, Edge Intelligence.
报告人
Lourdu Joanna J
ASSISTANT PROFESSOR India； Trichy；J.J. College of Engineering and Technology稿件作者
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
IEEE Madras Section
联系方式
- ICon INDIA 2026
- ic******@usssociety.org
发表评论