Abstract:This paper proposes a novel intelligent human activity recognition (HAR) framework based on a new design of Federated Split Learning (FSL) with Differential Privacy (DP) over edge networks. Our FSL-DP framework leverages both accelerometer and gyroscope data, achieving significant improvements in HAR accuracy. The evaluation includes a detailed comparison between traditional Federated Learning (FL) and our FSL framework, showing that the FSL framework outperforms FL models in both accuracy and loss metrics. Additionally, we examine the privacy-performance trade-off under different data settings in the DP mechanism, highlighting the balance between privacy guarantees and model accuracy. The results also indicate that our FSL framework achieves faster communication times per training round compared to traditional FL, further emphasizing its efficiency and effectiveness. This work provides valuable insight and a novel framework which was tested on a real-life dataset.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to improve the accuracy and training efficiency of the model while protecting user data privacy in human activity recognition (HAR). Specifically, the paper proposes a new framework based on Federated Split Learning (FSL) combined with Differential Privacy (DP), aiming to solve the following problems: 1. **Data privacy protection**: Traditional centralized learning methods require uploading users' data to a central server for training, which may lead to the leakage of user data. By using FSL and DP, model training can be carried out without sharing the original data, thus protecting the privacy of user data. 2. **Model performance improvement**: Compared with traditional Federated Learning (FL), FSL reduces the computational burden of resource - limited devices by partially performing forward propagation and backward propagation on network edge devices, and improves the training speed and accuracy of the model. 3. **Communication efficiency optimization**: FSL reduces the communication delay by reducing the amount of data that needs to be transmitted, especially on resource - limited Internet of Things (IoT) devices, and this advantage is more obvious. 4. **Multi - sensor data fusion**: The paper explores the impact of using accelerometer and gyroscope data simultaneously on HAR performance, and finds that combining the data of these two sensors can significantly improve the accuracy and robustness of the model. ### Main contributions of the paper 1. **Proposing the FSL algorithm**: In the edge computing environment, collaborative privacy - enhanced HAR is achieved through the FSL algorithm. Edge Devices (EDs) are responsible for part of the forward propagation and backward propagation, and the remaining computational tasks are completed by the edge server, thereby reducing the computational burden of resource - limited devices. 2. **Integrating the differential privacy mechanism**: In order to further protect the data privacy of the intermediate activation layer, the paper introduces the Local Rényi Differential Privacy (LRDP) mechanism of Gaussian noise to ensure the security of shared information. 3. **Experimentally proving effectiveness**: Through experiments on real HAR datasets, the effectiveness of the proposed FSL method is verified. The experimental results show that this method significantly improves the training performance of the model while reducing the training delay, and under different training settings, the impact of the differential privacy mechanism on the model performance is also within a controllable range. ### Conclusion The paper proposes a novel FSL framework, which combines the advantages of federated learning and split learning, and enhances data privacy protection through the differential privacy mechanism. The experimental results show that this framework performs excellently in improving the accuracy and training efficiency of HAR models, and also has significant advantages in communication time and privacy protection. These achievements provide an important reference for future personalized activity recognition systems in the Internet of Things and wearable technologies.

Federated Split Learning for Human Activity Recognition with Differential Privacy

Privacy in Multimodal Federated Human Activity Recognition

2D Federated Learning for Personalized Human Activity Recognition in Cyber-Physical-Social Systems

FedDL: Federated Learning via Dynamic Layer Sharing for Human Activity Recognition

A federated learning system with enhanced feature extraction for human activity recognition

CDFL: Efficient Federated Human Activity Recognition using Contrastive Learning and Deep Clustering

Performance-Enhanced Federated Learning with Differential Privacy for Internet of Things

Personalized Federated Human Activity Recognition through Semi-supervised Learning and Enhanced Representation

Privacy-Preserving Federated Learning with Differentially Private Hyperdimensional Computing

Using Decentralized Aggregation for Federated Learning with Differential Privacy

A Two-Stage Differential Privacy Scheme for Federated Learning Based on Edge Intelligence

Hierarchical Clustering-based Personalized Federated Learning for Robust and Fair Human Activity Recognition

FedHAR: Semi-Supervised Online Learning for Personalized Federated Human Activity Recognition

Democratizing Federated WiFi-based Human Activity Recognition Using Hypothesis Transfer

Federated Unlearning for Human Activity Recognition

Towards Adaptive Asynchronous Federated Learning for Human Activity Recognition

HFL-DP: Hierarchical Federated Learning with Differential Privacy

A Profile Similarity-Based Personalized Federated Learning Method for Wearable Sensor-Based Human Activity Recognition

Differential Privacy-enabled Federated Learning for Sensitive Health Data

ProtoHAR: Prototype Guided Personalized Federated Learning for Human Activity Recognition

Federated Learning with Heterogeneous Labels and Models for Mobile Activity Monitoring