Investigating differentially private learning for document key information extraction

Summary

Researchers conducted the first systematic study of privacy-preserving training for multimodal document key information extraction (KIE) models, comparing differential privacy (DP), federated learning (FL), and DP federated learning (DP-FL) across six benchmark datasets. They found that DP training causes substantial performance loss, which can be largely recovered through pretrained document foundation models, DP-suited hyperparameter tuning, and task-specific pretraining, retaining 77–93% of non-private baseline performance at ε ∈ {8, 20}. Non-private KIE models were shown to be highly vulnerable to membership inference attacks, which FedAVG and DP substantially mitigate.

Mitigation

The source recommends training with differential privacy, federated learning, or DP federated learning, using pretrained document foundation models, DP-suited hyperparameter tuning, and task-specific pretraining aligned to the private data distribution, and it provides practical guidelines for privacy-utility trade-offs.

Related items

• InfoA Systematic Review of Privacy-Preserving Approaches in Deep Learning ApplicationsSimilar attack · OpenAlex (peer-reviewed AI security)
• InfoCoRTeC: A Differentially Private Release for Synthetic Tabular Data, Decodable by Any GeneratorSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoTiming is Everything: Inferring Temporal Membership in Continually Updated ModelsSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoProbing Local Conditional Fit for Diffusion Model Membership InferenceSimilar attack · Arxiv (cs.RO + cs.CV security)
• InfoFrom Training Exposure to Membership Evidence: A Controlled Study of OLMo 2Similar attack · Arxiv (cs.CR + cs.CL + cs.LG)