Abstract:This paper proposes an energy-efficient federated learning method and its application in human activity monitoring and recognition. In the proposed approach, the device that needs a model for an application requests its nearby devices for collaboration. The nearby devices that accept the request, send their model updates to the requesting device. The device receives the model updates from the collaborators and performs aggregation to build its model. As mobile devices have limited battery life, the number of rounds is decided based on the desired accuracy level and battery level of the requesting device. The performance of the proposed approach is evaluated with respect to prediction accuracy, training time, training energy consumption of the device, and response time. We have used two different datasets for performance evaluation. The first dataset contains different types of physical activities and the respective calorie burn. The second dataset is a human activity recognition dataset that considers six types of physical activities. The experimental results show that using the proposed method the training time and training energy consumption of the device are reduced by approximately 59% and 19% for the first and second datasets respectively, than the decentralized federated learning approach, while using LSTM as the underlying data analysis model. The results also present that the proposed method reduces the training time and energy consumption by approximately 55% and 72% for the first and second datasets respectively, than the decentralized federated learning approach while using MLP as the underlying data analysis model.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to perform Human Activity Recognition (HAR) through Federated Learning (FL) in a resource - constrained environment with limited battery life of mobile devices, while ensuring the accuracy of the model, reducing training time and energy consumption. Specifically: 1. **Network Connection and Battery Life Limitations**: Since mobile devices (such as smart phones) may not always have a stable network connection and have limited battery life, traditional centralized or decentralized federated learning methods may not work effectively in such an environment. 2. **Personalization Requirements and Privacy Protection**: Many applications (such as health management applications, activity monitoring applications, etc.) need to process users' personal data, which may be sensitive. To protect users' privacy, data cannot be uploaded to the cloud for analysis. 3. **Efficient Collaborative Learning**: The paper proposes an Opportunistic Federated Learning method (EnFed). When a device needs to initialize or update the model, it can collaborate through other nearby devices without relying on cloud servers or forming a complete decentralized federated learning network. This not only reduces the dependence on network connections but also reduces energy consumption and improves training efficiency. 4. **Optimization Objectives**: Minimize training time and energy consumption as much as possible while meeting the required accuracy. The paper mentions that by selecting appropriate parameters (such as the number of rounds \(R\), batch size \(BA\), etc.), resource consumption can be minimized while ensuring the performance of the model. ### Formula Summary - **Objective Function**: - Minimize training time: \[ T_{MA}=\min(T(MA)) \] - Minimize energy consumption: \[ E_{MA}=\min(E(MA)) \] - Meet the required prediction accuracy: \[ \text{Accuracy}_{MA}\geq A_A \] - **Calculate Energy Consumption**: \[ E_{\text{comp}}=(T_{\text{init}}\cdot E_{ci})+(T_{\text{loc}}\cdot E_{cl})+(T_{\text{agg}}\cdot E_{ca}) \] \[ E_{\text{comm}}=(T_{\text{com1}}\cdot E_s)+(T_{\text{com2}}\cdot E_r) \] - **Total Energy Consumption**: \[ E_{MA}=E_{\text{comp}}+E_{\text{comm}} \] - **Loss Function**: \[ L(w_{MA})=\frac{1}{N_c}\sum_{j = 1}^{N_c}L(w_j^A) \] ### Experimental Results The paper uses two different datasets for experimental evaluation. The results show that the EnFed method has significant improvements in training time and energy consumption compared with traditional decentralized federated learning methods. Specifically: - For the first dataset (involving different types of physical activities and their calorie consumption), EnFed achieved a prediction accuracy of about 96.7% with 2 - 5 collaborators, and the training time and energy consumption were reduced by about 59% and 19% respectively. - For the second dataset (involving six types of physical activities), EnFed achieved a maximum accuracy of 98.12% with 5 collaborators, and the training time and energy consumption were reduced by about 55% and 72% respectively. In conclusion, this paper aims to address the challenges in a resource - constrained environment by proposing a new federated learning method, especially for the battery life and network connection problems of mobile devices, so as to achieve efficient and energy - saving human activity recognition.

EnFed: An Energy-aware Opportunistic Federated Learning in Resource Constrained Environments for Human Activity Recognition

Federated Learning with Heterogeneous Labels and Models for Mobile Activity Monitoring

Energy-Aware Edge Association for Cluster-based Personalized Federated Learning

Energy Prediction using Federated Learning

Evaluation and comparison of federated learning algorithms for Human Activity Recognition on smartphones

FedAAR: A Novel Federated Learning Framework for Animal Activity Recognition with Wearable Sensors

Towards Energy-efficient Resource Allocation for Federated Learning in Mobile Edge Computing

Learn More by Using Less: Distributed Learning with Energy-Constrained Devices

DEAL: Decremental Energy-Aware Learning in a Federated System

Energy-efficient Federated Learning Via Stabilization-aware On-device Update Scaling.

Towards Energy-Aware Federated Learning on Battery-Powered Clients

Federated Action Recognition on Heterogeneous Embedded Devices

FedOpenHAR: Federated Multi-Task Transfer Learning for Sensor-Based Human Activity Recognition

Energy-Aware Federated Learning with Distributed User Sampling and Multichannel ALOHA

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

Federated learning for energy constrained devices: a systematic mapping study

Energy-Aware Heterogeneous Federated Learning via Approximate DNN Accelerators

Federated Meta-Learning with Attention for Diversity-Aware Human Activity Recognition

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

Resource Aware Clustering for Tackling the Heterogeneity of Participants in Federated Learning

Meta-HAR: Federated Representation Learning for Human Activity Recognition