Research map / Image generation / GANs
Influence Estimation for Generative Adversarial Networks
Naoyuki Terashita, Hiroki Ohashi, Yuichi Nonaka, Takashi Kanemaru
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 estimated influence on ALL to select and remove training instances | FCGAN · 2D-Normal | test ALL: +0.44 (0.50) | Improved |
| Remove too many instances selected by influence on ALL (n_h=20.0k) | FCGAN · 2D-Normal | test ALL: -1.67 (2.33) | Hurt |
| Select instances using influence on discriminator loss rather than the GAN evaluation metric | FCGAN · 2D-Normal | test ALL: +0.19 (0.19) | Improved |
| Use estimated influence on IS to select and remove training instances | DCGAN · MNIST | test IS: +0.14 (0.14) | Improved |
| Remove too many instances selected by influence on IS (n_h=45.0k) | DCGAN · MNIST | test IS: -0.07 (0.24) | Hurt |
| Use estimated influence on FID to select and remove training instances | DCGAN · MNIST | test IS: +0.10 (0.13) | Improved |
| Use estimated influence on FID to select and remove training instances | DCGAN · MNIST | test FID: -0.38 (0.37) | Improved |
| Use estimated influence on IS to select and remove training instances | DCGAN · MNIST | test FID: -0.36 (0.30) | Improved |
| Select instances using discriminator-loss influence instead of the GAN evaluation metric | DCGAN · MNIST | test FID: -0.15 (0.14) | Improved |
| Remove too many instances selected by influence on FID (n_h=45.0k) | DCGAN · MNIST | test FID: +0.23 (0.60) | Hurt |
| Increase the number of epochs traced back in influence estimation (ablation) | GAN influence estimator · 2D-Normal and MNIST | estimation accuracy | Hurt |
Closest papers by meaning
- Instance Selection for GANs (NeurIPS 2020 · similarity 0.87)
- Learning from Bad Data via Generation (NeurIPS 2019 · similarity 0.87)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.87)
- Quantitatively Evaluating GANs With Divergences Proposed for Training (ICLR 2018 · similarity 0.86)
- A Domain Agnostic Measure for Monitoring and Evaluating GANs (NeurIPS 2019 · similarity 0.86)
- Data-Efficient Instance Generation from Instance Discrimination (NeurIPS 2021 · similarity 0.86)
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.
