Private Contrastive Evolution

Private Contrastive Evolution (PCEvolve) is a novel API-assisted algorithm jointly proposed by Shanghai Jiao Tong University and Tsinghua University on June 4, 2025. This method mines the inherent inter-category contrast relationship in a small number of private data in an iterative process, rather than relying solely on a single data point.PCEvolve: Private Contrastive Evolution for Synthetic Dataset Generation via Few-Shot Private Data and Generative APIs".

Extensive experimental results demonstrate that PCEvolve outperforms PE and other API-based baseline methods across multiple metrics. These results highlight the potential of combining private data with API access for quality assessment, enabling the generation of high-quality differentially private synthetic images and paving the way for easier-to-use and more efficient privacy-preserving generative API applications.