AI summaryⓘ
The authors studied a way to train AI models called federated learning, where data stays on users' devices and only model updates are shared with a central server. They compared how this method works against traditional centralized and local training approaches. They tested their ideas specifically for recognizing human activities and looked at how well models balance being good for everyone versus being personalized for each user. Their results show that federated learning can personalize models while still being generally accurate, but this advantage becomes less clear when the users' data changes a lot. Overall, the authors highlight challenges in managing the tradeoff between personalization and generalization in federated learning.
Federated LearningCentralized LearningLocal LearningModel PersonalizationModel GeneralizationFedAvg AlgorithmHuman Activity Recognition (HAR)Data DistributionPrivacyDistributed Computing
Authors
Andrea De Luna, Susanna Peretti, Chiara Contoli, Alessandro Bogliolo
Abstract
The federated learning (FL) paradigm fosters distributed pervasive computing combined with artificial intelligence techniques, allowing for optimized data usage and improved mitigation of privacy concerns. Indeed, model training occurs on the client's local devices, and model parameters are subsequently shared with a centralized server. However, there is a need to find a tradeoff between models' personalization and generalization capabilities. In this paper, we design and implement several testing scenarios devoted to evaluating and comparing the centralized, local, and federated paradigm performances. We also design and implement a scenario that emulates a change in clients' data. We then present experimental results of the FedAvg algorithm applied to the Human Activity Recognition (HAR) domain to understand the trade-off between personalized and generalized accuracy. Results show that, although FedAvg confirms a higher degree of personalization capabilities while keeping a high degree of generalization with respect to the traditional centralized learning, this result is not so obvious under stressful conditions, such as when varying class distribution over clients.