End to End Collaborative Synthetic Data Generation

TL;DR

The paper tackles end-to-end privacy-preserving collaborative synthetic data generation across multiple data custodians, aiming to publish DP-compliant synthetic data without exposing raw inputs.It introduces an MPC-based framework (DP-in-MPC) that orchestrates privacy-preserving preprocessing, SDG model training (hyperparameter tuning via k-fold cross-validation), and evaluation, culminating in publishable synthetic data with bounded privacy loss $$(\epsilon,\delta)$$.The authors instantiate the framework for a leukemia genomic use case using Private-PGM and MPC protocols for quantile binning, marginal computation, and secure evaluation, demonstrating feasibility and outlining performance in terms of runtime and privacy budgets.Key contributions include a modular end-to-end pipeline that preserves input privacy via MPC, preserves output privacy via DP, and enables multi-stage SDG (including hyperparameter tuning) across silos, with an empirical evaluation on genomic data and clear discussion of deployment challenges.

Abstract

The success of AI is based on the availability of data to train models. While in some cases a single data custodian may have sufficient data to enable AI, often multiple custodians need to collaborate to reach a cumulative size required for meaningful AI research. The latter is, for example, often the case for rare diseases, with each clinical site having data for only a small number of patients. Recent algorithms for federated synthetic data generation are an important step towards collaborative, privacy-preserving data sharing. Existing techniques, however, focus exclusively on synthesizer training, assuming that the training data is already preprocessed and that the desired synthetic data can be delivered in one shot, without any hyperparameter tuning. In this paper, we propose an end-to-end collaborative framework for publishing of synthetic data that accounts for privacy-preserving preprocessing as well as evaluation. We instantiate this framework with Secure Multiparty Computation (MPC) protocols and evaluate it in a use case for privacy-preserving publishing of synthetic genomic data for leukemia.

Figures (1)

• Figure 1: Proposed framework for privacy-preserving publishing of synthetic data