Locally Private Parametric Methods for Change-Point Detection

Researchers have published a new paper establishing the statistical cost of local differential privacy in change-point detection algorithms. They developed two private algorithms and proved that privacy inevitably degrades performance compared to non-private benchmarks. A key theoretical finding shows that binary input distributions achieve optimality for strong data processing inequalities. This work provides concrete metrics for the privacy-accuracy tradeoff, validated across 20 figures and 43 pages of analysis.