Reinforcement-Guided Synthetic Data Generation for Privacy-Sensitive Identity Recognition

2026-04-09

AI Summary

This work addresses the challenge of limited data availability in privacy-sensitive scenarios, which severely constrains the performance of generative models. To overcome this, the authors propose a reinforcement-guided synthetic data generation framework that first aligns a pretrained generator to the target domain via cold-start adaptation. They then introduce a multi-objective reinforcement learning reward mechanism that jointly optimizes semantic consistency, diversity, and expressive richness, complemented by a dynamic high-value sample selection strategy to enhance downstream task performance. Notably, this is the first approach to apply reinforcement learning–based guidance to synthetic data generation for privacy-sensitive identity recognition. The method achieves significant improvements in generation fidelity and classification accuracy across multiple benchmarks and demonstrates strong generalization to novel classes under few-shot settings.

Abstract

High-fidelity generative models are increasingly needed in privacy-sensitive scenarios, where access to data is severely restricted due to regulatory and copyright constraints. This scarcity hampers model development--ironically, in settings where generative models are most needed to compensate for the lack of data. This creates a self-reinforcing challenge: limited data leads to poor generative models, which in turn fail to mitigate data scarcity. To break this cycle, we propose a reinforcement-guided synthetic data generation framework that adapts general-domain generative priors to privacy-sensitive identity recognition tasks. We first perform a cold-start adaptation to align a pretrained generator with the target domain, establishing semantic relevance and initial fidelity. Building on this foundation, we introduce a multi-objective reward that jointly optimizes semantic consistency, coverage diversity, and expression richness, guiding the generator to produce both realistic and task-effective samples. During downstream training, a dynamic sample selection mechanism further prioritizes high-utility synthetic samples, enabling adaptive data scaling and improved domain alignment. Extensive experiments on benchmark datasets demonstrate that our framework significantly improves both generation fidelity and classification accuracy, while also exhibiting strong generalization to novel categories in small-data regimes.

Problem

• privacy-sensitive
• identity recognition
• data scarcity
• synthetic data generation
• generative models

Innovation

• reinforcement-guided generation
• synthetic data
• privacy-sensitive recognition
• multi-objective reward
• dynamic sample selection