Research map / Image generation / GANs
Generative Adversarial Nets for Robust Scatter Estimation: A Proper Scoring Rule Perspective
Chao Gao, Yuan Yao, Weizhi Zhu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use JS-GAN instead of Tyler's M-estimator | JS-GAN G1 · t-distribution v=1 | spectral norm error: 372.9637 -> 0.2808 | Improved |
| use JS-GAN instead of dimension halving | JS-GAN G1 · t-distribution v=4 | spectral norm error: 1.2834 -> 0.2751 | Improved |
| increase contamination distance s | JS-GAN G1 · Dirac contamination | spectral norm error: 0.2000 -> 0.1600 | Improved |
| increase contamination distance s | Kendall's tau · Dirac contamination | spectral norm error: 3.8098 -> 92.1079 | Hurt |
| increase contamination proportion epsilon | Tyler's M-estimator, Kendall's tau, MVE · Gaussian contamination, p=100 | spectral norm error: 4.1114 -> 39.3082 | Hurt |
| increase contamination proportion epsilon | JS-GAN · p=100 | spectral norm error: 0.1438 -> 0.1620 | No clear effect |
| increase sample size n from 5,000 to 50,000 | JS-GAN G1 · N(5,5I) contamination | spectral norm error: 1.0186 -> 0.1547 | Improved |
| increase sample size n | Tyler's M-estimator · N(5,5I) contamination | spectral norm error: 38.6494 -> 39.2105 | No clear effect |
| increase dimension p from 10 to 200 | JS-GAN G1 · N(5,5I) contamination | spectral norm error: 0.0498 -> 0.2154 | Hurt |
| learn the radial distribution with extra network (G2 vs G1) (ablation) | JS-GAN · Simultaneous estimation | spectral norm error: 0.2808 -> 0.3350 | Hurt |
| add location estimation (G3/G4 vs G1/G2) (ablation) | JS-GAN · Simultaneous estimation | scatter error: 0.1615 -> 0.1537 | No clear effect |
| use asymmetric Beta-GAN(1,0.5) instead of JS-GAN (ablation) | Beta-GAN · N(0,I) with N(5,5I), p=100 | spectral norm error: 0.1188 -> 0.1040 | Improved |
Closest papers by meaning
- ROBUST ESTIMATION VIA GENERATIVE ADVERSARIAL NETWORKS (ICLR 2019 · similarity 0.88)
- Tractable and Near-Optimal Adversarial Algorithms for Robust Estimation in Contaminated Gaussian Models (JMLR 2023 · similarity 0.87)
- A high-quality robust diffusion framework for corrupted dataset (ECCV 2024 · similarity 0.87)
- On the Statistical Properties of Generative Adversarial Models for Low Intrinsic Data Dimension (JMLR 2025 · similarity 0.86)
- Robust Learning Meets Generative Models: Can Proxy Distributions Improve Adversarial Robustness? (ICLR 2022 · similarity 0.86)
- SMaRt: Improving GANs with Score Matching Regularity (ICML 2024 · similarity 0.85)
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.
