PMF: A Privacy-preserving Human Mobility Prediction Framework via Federated Learning
summary
Proposes a mobility predictor that keeps raw trajectories on-device and only exchanges model weights, combining federated learning with differential privacy noise on the location embeddings and a two-stage embedding freeze training schedule plus a local personal adaptor for fine-tuning. Implements an LSTM with embedding tables, evaluates on three real-world datasets (two check-in, one dense trajectory), and reports competitive accuracy versus centralized joint models, better than per-user models, and more stable accuracy as privacy strength increases.
AI-written summary · about
related talks
28mA Privacy Enhancing Federated Learning Framework for Multi-Party Distributed ..
14mSecure FLOATING - Scalable Federated Learning Framework for Real-time Trust in...
46mFrom Health Data to Location Traces: Addressing Spatiotemporal Correlations in Differential Privacy
44mPersonalized Federated Training of Diffusion Models with Privacy Guarantees
27mCreate Value and Insight Through Privacy-preserving Data Collaboration [...]
06mPREFER: Point-of-interest REcommendation with efficiency and privacy-preservation via Federated ...
34mDecentralized Machine Learning
41mFederated Learning: Machine Learning on Decentralized Data