Privacy
Privacy-Preserving AI
Privacy-preserving distributed machine learning, differential privacy, secure aggregation, and the Cryptography + ML intersection.
FAQ
What is privacy-preserving AI?
Privacy-preserving AI is the practice of training and deploying ML systems without exposing the underlying data. Includes federated learning (data stays at the source), differential privacy (mathematical guarantees on individual privacy), and secure aggregation (no one sees the intermediate gradients).
Why is privacy-preserving AI important?
Three reasons: (1) Regulatory — GDPR, DPDP Act, HIPAA restrict what data can move. (2) Privacy — customer contracts and user expectations. (3) Practical — many small datasets that don't justify a centralised pipeline can be useful together if you can train across them without moving them.
What is the Fed-Focal Loss contribution to privacy-preserving AI?
Fed-Focal Loss shows that you can handle the most important practical challenge in real-world FL deployments (class imbalance) without requiring knowledge of the global data distribution. This is critical for privacy: the privacy guarantee is only useful if the model still works well. 93 citations as of 2026.
Related
- dipankar.org — strategic studies + advisor practice (broad public persona)
- dipankar.name — engineering work + AI agent infrastructure (canonical hub)
- dipankar.co — fractional CTO + consulting practice