Enhancing privacy-preserving AI in mobile keyboards: Federated learning and differential privacy approaches
Abstract
Machine learning models in mobile keyboards enhance user experience through emoji prediction, auto-correction, and next-word suggestions. However, centralized data aggregation raises privacy concerns. Federated Learning (FL) offers a decentralized alternative, enabling on-device training without uploading private data. Google&s;s Gboard uses FL with Federated Averaging (FedAvg) and Differential Privacy (DP-FTRL) to ensure privacy. Secure Aggregation (SecAgg) encrypts updates, while Distributed DP enhances safeguards. FL-based models outperform traditional ones in accuracy, with top-1 recall improvements exceeding 16%, ensuring privacy without compromising utility.
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.