JayQuery: Privacy-Preserving Student Success Prediction
JayQuery was ADAL’s SCARP 2026 project on privacy-preserving student success prediction with federated learning.
Colleges are data-rich but insight-poor: FERPA regulations prevent offices such as the Registrar, Financial Aid and the learning management system from merging their student records into a single model. JayQuery uses vertical federated learning (VFL): separate data silos collaborate to train a dropout-risk predictor without ever sharing raw records. The silos align their records through a private set intersection (PSI) handshake, each trains a local “bottom” model, and a federated learning server combines their embeddings in a “top” model protected with differential-privacy noise.
Team
- Alexander Pellet, system architect and ML engineer
- Wyatt Wilfong, privacy and security lead
- Dr. Peilong Li, faculty mentor