Skip to content

fairlearn-fhe

Drop-in encrypted Fairlearn metrics. Fairlearn-compatible API for the supported metric subset; ciphertext arithmetic over CKKS via TenSEAL or OpenFHE.

fairlearn-fhe is an early-stage project maintained at BAder82t/fairlearn-fhe.

# plaintext (existing Fairlearn workflow)
from fairlearn.metrics import demographic_parity_difference
disp = demographic_parity_difference(y_true, y_pred, sensitive_features=A)

# encrypted (one import change)
from fairlearn_fhe import build_context, encrypt
from fairlearn_fhe.metrics import demographic_parity_difference

ctx = build_context()
y_pred_enc = encrypt(ctx, y_pred)
disp = demographic_parity_difference(y_true, y_pred_enc, sensitive_features=A)

disp is numerically equivalent to the plaintext result within CKKS noise tolerance (<1e-4 abs error in default settings; typically ~1e-7 on TenSEAL, ~1e-10 on OpenFHE).

Why

Fairness audits often require the auditor to see labels and group-level denominators while the vendor keeps model predictions confidential. Fairlearn (the JMLR'23 standard fairness library) has no FHE plugin. fairlearn-fhe ports the canonical metric set to CKKS without changing the call signature.

What is shipped

  • 12 canonical metrics (selection_rate, mean_prediction, true/false positive/negative rate, demographic parity diff/ratio, equalized odds diff/ratio, equal opportunity diff/ratio).
  • MetricFrame and make_derived_metric with the same .by_group, .overall, .difference(), .ratio(), .group_min(), .group_max() accessors.
  • Two trust models: encrypted y_pred only (Mode A); encrypted y_pred + sensitive features (Mode B).
  • Two CKKS backends: TenSEAL (pip) and OpenFHE (native backend, opt-in).
  • A canonical MetricEnvelope with parameter-set hash, observed depth, and op counts for regulator logging.

Trust posture

  • y_true is plaintext — it is the auditor's ground truth.
  • y_pred is encrypted (always).
  • sensitive_features are plaintext (Mode A) or encrypted with revealed group counts (Mode B).
  • Group-size denominators (and TPR/FPR positive/negative counts) are auditor-public metadata. There is no fairness signal without them.

See Trust models for the full posture.

Install

pip install fairlearn-fhe          # tenseal backend
pip install fairlearn-fhe[openfhe] # add openfhe backend (requires C++ build)

License

Apache-2.0. Compatible with Fairlearn (MIT).