Authors: EMBL

Affiliation: EMBL

Description: Baseline method submission,
see https://github.com/PMBio/Health-Privacy-Challenge for details

method: MAMA-MIA on RNASeq Data2025-03-16

Authors: Steven Golob, Sikha Pentyala, Carter Bennet, Terri Bell, Martine De Cock

Affiliation: University of Washington Tacoma

Description: We run MAMA-MIA [1] on RNASeq Data. MAMA-MIA is tuned to MST [2] data generation to estimate relevant attack information (i.e. focal points) to perform the attack.

[1] Privacy Vulnerabilities in Marginals-based Synthetic Data.
S. Golob, S. Pentyala, A. Maratkhan, M. De Cock. IEEE Secure and Trustworthy Machine Learning Conference (SaTML), 2025

[2] Winning the NIST Contest: A scalable and general approach to differentially private synthetic data. McKenna, Ryan, Daniel Sheldon, and Gerome Miklau. Journal of Privacy and Confidentiality 11 (3)

Authors: Sikha Pentyala, Steven Golob, Carter Bennet, Terri Bell, Martine De Cock

Affiliation: University of Washington Tacoma

Description: Our baseline submission leverages prior information that 80% of the target dataset is in the training dataset. This information is enhanced with distance metrics between the synthetic and target datasets.

Ranking Table

Description Paper Source Code
Set1Set2Set3
DateMethodAUCPR_AUCTPR@FPR=0.01TPR@FPR=0.1AUCPR_AUCTPR@FPR=0.01TPR@FPR=0.1AUCPR_AUCTPR@FPR=0.01TPR@FPR=0.1
2025-03-21Baseline (Monte Carlo, Hilprecht et al., 2019)47.96%78.49%0.46%8.50%50.42%79.69%0.57%9.07%52.00%82.15%2.18%18.48%
2025-03-16MAMA-MIA on RNASeq Data52.88%81.82%2.07%12.86%47.18%78.13%0.57%7.12%52.78%81.91%1.38%12.06%
2025-03-15MIA with Distance-Based Metrics and Prior Knowledge53.09%80.75%0.23%9.76%51.44%80.83%2.53%10.33%51.69%82.13%3.67%13.89%

Ranking Graphic

Ranking Graphic

Ranking Graphic

Ranking Graphic

Ranking Graphic

Ranking Graphic