Re-Identification vs Anonymization Strength
2026-02-08 — Exploring how increasing k-anonymity affects data utility and the attacker’s ability to re-identify records.
Recent posts
Using geometry to choose embeddings
2025-11-11 — Empirical evaluation of local geometry in vector embeddings across models and corpora.
Model Equivalence using Z3
2025-11-07 — Using Z3 to prove two ML models are logically equivalent — or extract the exact counterexample where they diverge.
Quantifying Information Loss
2025-10-28 — A quick experiment linking Laplace noise and data resolution, showing how privacy and precision trade off
Model based simulations
2023-06-09 — Bayesian decision making applied to sales opportunities
Noise, Stability, and Calibration
2021-05-01 — How models behave when data gets messy