AugGen: Synthetic Augmentation Can Improve Discriminative Models

2025-03-14

AI Summary

To address the scarcity of labeled data in privacy-sensitive scenarios and the reliance of existing synthetic data methods on external datasets or pre-trained models, this paper proposes a fully self-contained, closed-loop synthetic augmentation paradigm. Our approach trains a conditional generative model exclusively on the target dataset and directly samples synthetic instances from it—requiring no external data or pre-trained weights. By jointly optimizing the generator and discriminator, the method achieves endogenous data augmentation. Evaluated on the IJB-C and IJB-B face recognition benchmarks, it improves identification accuracy by 1–12% over real-data-only baselines and outperforms state-of-the-art synthetic-data methods. Notably, it is the first to demonstrate that synthetic augmentation alone can surpass performance gains achieved through mainstream network architecture improvements. This work establishes an efficient, privacy-compliant pathway for enhancing model performance under strict data governance constraints.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Privacy
• Natural Language Processing: Generation

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Security and Privacy: Data transparency and provenance

Abstract

The increasing dependence on large-scale datasets in machine learning introduces significant privacy and ethical challenges. Synthetic data generation offers a promising solution; however, most current methods rely on external datasets or pre-trained models, which add complexity and escalate resource demands. In this work, we introduce a novel self-contained synthetic augmentation technique that strategically samples from a conditional generative model trained exclusively on the target dataset. This approach eliminates the need for auxiliary data sources. Applied to face recognition datasets, our method achieves 1--12% performance improvements on the IJB-C and IJB-B benchmarks. It outperforms models trained solely on real data and exceeds the performance of state-of-the-art synthetic data generation baselines. Notably, these enhancements often surpass those achieved through architectural improvements, underscoring the significant impact of synthetic augmentation in data-scarce environments. These findings demonstrate that carefully integrated synthetic data not only addresses privacy and resource constraints but also substantially boosts model performance. Project page https://parsa-ra.github.io/auggen

Problem

• Addresses privacy and ethical challenges in machine learning datasets.
• Eliminates need for external datasets in synthetic data generation.
• Improves model performance in data-scarce environments using synthetic augmentation.

Innovation

• Self-contained synthetic augmentation technique
• Conditional generative model on target dataset
• Eliminates need for auxiliary data sources