Evaluation of Thermal Stress on IoT-based Federated Learning
Disciplines
Other Computer Engineering
Abstract (300 words maximum)
Federated Learning is a novel paradigm allowing the training of a global machine-learning model on distributed devices. It shares model parameters or gradients instead of the private raw data during the entire model training process. While Federated Learning enables machine learning processes to take place collaboratively on the Internet of Things (IoT) devices, compared to data centers, IoT devices are with limited resource budgets and typically have less security protection. This makes IoT devices more vulnerable to potential cyber-attacks. Current research on the evaluation of Federated Learning is mainly based on the simulation of multi-client or multi-processed Federated Learning Systems deployed on a single machine or device. However, when it comes to understanding the performance of Federated Learning Systems under cyber-attacks, there is a gap between simulated Federated Learning Systems and real-world distributed Federated Learning Systems on low-power IoT devices. In this paper, we are among the first to evaluate the performance of Federated Learning Systems under thermal stress on real-world IoT-based distributed systems. We conducted comprehensive experiments using the CIFAR-10 dataset and measured various performance metrics including training time, CPU and GPU utilization rate, temperature, and power consumption. The experimental results demonstrate that thermal stress is effective in IoT-based Federated Learning systems. In general, thermal stress can negatively impact the entire global model and cause device performance to degrade even when even a small ratio of IoT devices is being impacted. The negative impacts are amplified while the ratio of IoT devices impacted goes up.
Academic department under which the project should be listed
CCSE - Information Technology
Primary Investigator (PI) Name
Liang Zhao
Additional Faculty
Bobin Deng, Computer Science, bdeng2@kennesaw.edu
Shaoen Wu, Information Technology, swu10@kennesaw.edu
Evaluation of Thermal Stress on IoT-based Federated Learning
Federated Learning is a novel paradigm allowing the training of a global machine-learning model on distributed devices. It shares model parameters or gradients instead of the private raw data during the entire model training process. While Federated Learning enables machine learning processes to take place collaboratively on the Internet of Things (IoT) devices, compared to data centers, IoT devices are with limited resource budgets and typically have less security protection. This makes IoT devices more vulnerable to potential cyber-attacks. Current research on the evaluation of Federated Learning is mainly based on the simulation of multi-client or multi-processed Federated Learning Systems deployed on a single machine or device. However, when it comes to understanding the performance of Federated Learning Systems under cyber-attacks, there is a gap between simulated Federated Learning Systems and real-world distributed Federated Learning Systems on low-power IoT devices. In this paper, we are among the first to evaluate the performance of Federated Learning Systems under thermal stress on real-world IoT-based distributed systems. We conducted comprehensive experiments using the CIFAR-10 dataset and measured various performance metrics including training time, CPU and GPU utilization rate, temperature, and power consumption. The experimental results demonstrate that thermal stress is effective in IoT-based Federated Learning systems. In general, thermal stress can negatively impact the entire global model and cause device performance to degrade even when even a small ratio of IoT devices is being impacted. The negative impacts are amplified while the ratio of IoT devices impacted goes up.