Membership inference-guided training data extraction attack on diffusion models
Start here
Researchers propose MGE, a membership-inference-guided data extraction attack framework that can extract training data from pretrained diffusion models. The method models data extraction as a conditional generation task, with the conditioning signal coming from a membership inference classifier, and improves attack success rate and efficiency through adversarial gradients and parallel branch decoding. Experiments on image generation benchmarks show that MGE can extract hundreds of images, and the authors argue that privacy protection for diffusion models needs more attention.
Generated by AI from reports · Updated 2 days ago
Attention and developments
Attention, hourly
Now 29, -2% in 24 hours·Peak 31 (Nov 10 22:00)
Related research
This event's paper
- MGE: A Membership Guided Data Extraction Attack against Diffusion Models
arXiv · 2026-11-07 · Deep reading
Progress timeline
- arXiv: Jailbreaks, Prompt Injection, Poisoning & DefensesMGE: Membership-Inference-Guided Data Extraction Attack against Diffusion Models
The researchers propose MGE, a membership-inference-guided data extraction attack framework that can extract training data from pretrained diffusion models. The method models data extraction as a conditional generation task, with the conditioning signal coming from a membership inference classifier, and improves attack success rate and efficiency through adversarial gradients and parallel branch decoding. Experiments on image generation benchmarks show that MGE can extract hundreds of images, and the authors argue that privacy protection for diffusion models needs more attention.
Related discussion
Attention trend
New sources per day
- Nov 10: 1 new sources (1 outlets, 0 accounts)
- Nov 11: 0 new sources (0 outlets, 0 accounts)
- Nov 12: 0 new sources (0 outlets, 0 accounts)
- Nov 13: 0 new sources (0 outlets, 0 accounts)