Research map / Federated learning and privacy / Privacy attacks and secure learning

A Privacy-Preserving Framework for Generative Model-driven Synthetic Datasets

AAAI 2025 · Federated learning and privacy › Privacy attacks and secure learning · approach: Encryption-based private learning · 4 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Doctoral consortium-style proposal outlining a planned privacy-preserving, differentially private synthetic data framework for time-series, plus a survey and taxonomies of privacy/utility metrics for generative models.

Problem

Generative model-based synthetic data is vulnerable to privacy attacks (membership, attribute, model inference), and existing differential privacy approaches reduce utility and are hard to calibrate; standardized privacy/utility evaluation metrics are lacking, especially for time series.

Approach

Proposes an extensible framework for privacy-preserving synthetic time-series generation that integrates generative models, attacks and DP-based defenses (DPSGD), together with planned optimized metric suites to evaluate privacy/utility trade-offs.

Key results

• The framework was reported to show effectiveness with state-of-the-art models on downstream classification and regression tasks across different privacy budgets, though no numbers are given in the text.

Limitations

• Abstract-level doctoral proposal; the time-series framework evaluation is only summarized without reported figures
• Metrics suite optimization and privacy/utility trade-off analysis are listed as future work and not yet done

Experimental setup

Metrics: downstream classification performance downstream regression performance

Builds on: DPSGD GAN VAE

Closest papers by meaning

• Privacy Amplification Through Synthetic Data: Insights from Linear Regression (ICML 2025 · similarity 0.90)
• Does Training with Synthetic Data Truly Protect Privacy? (ICLR 2025 · similarity 0.89)
• PCEvolve: Private Contrastive Evolution for Synthetic Dataset Generation via Few-Shot Private Data and Generative APIs (ICML 2025 · similarity 0.89)
• Privacy Amplification Persists under Unlimited Synthetic Data Release (NeurIPS 2026 · similarity 0.89)
• CaPS: Collaborative and Private Synthetic Data Generation from Distributed Sources (ICML 2024 · similarity 0.88)
• Differentially Private Sum-Product Networks (ICML 2024 · similarity 0.88)