Research map / Image generation / Fast diffusion sampling
Generative Data Augmentation via Diffusion Distillation, Adversarial Alignment, and Importance Reweighting
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes an efficient generative data augmentation method that distills a diffusion model into a fast generator and corrects the synthetic-real distribution gap via adversarial alignment and importance reweighting.
Problem
Diffusion models are impractical for generative data augmentation because thousand-step sampling is slow and energy-intensive, and the distribution gap between synthetic and real data hurts downstream training.
Approach
Combines diffusion distillation for fast sampling, adversarial alignment to close the synthetic-real distribution gap, and importance reweighting of synthetic samples, so that generated augmentations are both cheap to produce and better matched to the real data distribution.
Closest papers by meaning
- GDA: Generalized Diffusion for Robust Test-time Adaptation (CVPR 2024 · similarity 0.88)
- Data Augmentation with Diffusion for Open-Set Semi-Supervised Learning (NeurIPS 2024 · similarity 0.88)
- Effective Data Augmentation With Diffusion Models (ICLR 2024 · similarity 0.87)
- Unified Generation, Reconstruction, and Representation: Generalized Diffusion with Adaptive Latent Encoding-Decoding (ICML 2024 · similarity 0.87)
- Instruct Where the Model Fails: Generative Data Augmentation via Guided Self-contrastive Fine-tuning (AAAI 2025 · similarity 0.87)
- REPA Works Until It Doesn’t: Early-Stopped, Holistic Alignment Supercharges Diffusion Training (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.