Floating-Point Fully Homomorphic Encryption (FPFHE)
A didactic website aligned with my thesis: motivation, theory, and hands-on examples.
What is FPFHE?
Floating-Point Fully Homomorphic Encryption (FPFHE) refers to FHE schemes and techniques that enable computation over floating-point–like data while it remains encrypted. Depending on the construction, “floating point” can mean either:
- Approximate arithmetic over real/complex numbers (e.g., CKKS), which is practical and fast for analytics and ML, but not bit-exact IEEE-754.
- Bit-level floating-point representations built on TFHE-style programmable bootstrapping, targeting IEEE-like behaviors (e.g., precision formats, overflow handling).
Why it matters
- Data confidentiality: compute on sensitive data without decrypting it on the server.
- Reduced trust: outsource computations to untrusted infrastructure with strong privacy guarantees.
- New deployment models: privacy-preserving analytics, ML inference, and secure collaboration.
Application areas
Privacy-preserving analytics
Aggregate statistics on encrypted datasets for regulated domains.
Secure ML inference
Run inference on encrypted features while protecting user inputs.
Healthcare & finance
Compute risk scores, signals, and models while preserving confidentiality.
Outsourced computation
Cloud execution without exposing plaintext data to the cloud provider.
Start here
The first implemented module is CKKS (approximate arithmetic over real numbers).
Go to CKKS