Differential Privacy: Exploring Federated Learning Privacy Issue to Improve Mobility Quality
Abstract
The increasing number of vehicles within cities has been stimulating intense research for smart tools to avoid traffic jams and improve mobility quality. To get the most accurate data, urban mobility applications consume data generated by users as they explain the mobility pattern of a studied region, entailing the meantime vulnerability of privacy of these latter. Thus, new techniques are considered to share the minimum information possible to train models and keep privacy. Federated Learning (FL) is one approach that exploits this concept, sharing only the gradients produced by local models trained at each device. Nonetheless, information can be learned during the training, exposing vulnerable content. In this context, this study aimed to validate the FL in a given application and check the data leakage in this decentralized training architecture through the differential privacy (DP) method, removing potential clients that make the architecture vulnerable. Finally, we ended up having a final FL model with 71% of accuracy and removing two processing units from the training process.
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.