Research map / Optimization / Implicit bias and generalization
Stabilizing Adversarial Nets with Prediction Methods
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Add prediction to both generator networks | Stacked GAN · CIFAR-10, learning rate 0.001 | Inception Score: 1.52 \pm 0.01 -> 7.96 \pm 0.11 | Helped |
| Add prediction to both generator networks | Stacked GAN · CIFAR-10, learning rate 0.0005 | Inception Score: 7.90 \pm 0.08 -> 8.13 \pm 0.09 | Helped |
| Add prediction to both generator networks | Stacked GAN · CIFAR-10, learning rate 0.0001 | Inception Score: 8.44 \pm 0.11 -> 8.55 \pm 0.12 | Helped |
| Add prediction to both generator networks | Stacked GAN · CIFAR-10, learning rate 0.0005, 40 epochs | Inception Score: 5.80 \pm 0.15 -> 8.10 \pm 0.10 | Helped |
| Add prediction to both generator networks | Stacked GAN · CIFAR-10, learning rate 0.001, 20 epochs | Inception Score: 1.42 \pm 0.01 -> 7.79 \pm 0.07 | Helped |
| Add prediction | DANN · Office, Amazon to Webcam | target domain accuracy: 73.4 -> 74.7 | Helped |
| Add prediction | DANN · Office, Webcam to Amazon | target domain accuracy: 51.6 -> 58.5 | Helped |
| Add prediction | DANN · Office, Dslr to Webcam | target domain accuracy: 95.5 -> 96.1 | Helped |
| Add prediction | DANN · Office, Webcam to Dslr | target domain accuracy: 99.4 -> 99.0 | Hurt |
| Add prediction | DANN · Office, Amazon to Dslr | target domain accuracy: 76.5 -> 73.5 | Hurt |
| Add prediction | DANN · Office, Dslr to Amazon | target domain accuracy: 51.7 -> 57.6 | Helped |
| Add a prediction step to alternating adversarial training | DCGAN · CIFAR-10, Adam learning rate 0.001, five random seeds | collapse frequency: all five random seeds -> only once | Helped |
Closest papers by meaning
- Stability and Generalization of Adversarial Training for Shallow Neural Networks with Smooth Activation (NeurIPS 2024 · similarity 0.89)
- Train faster, generalize better: Stability of stochastic gradient descent (ICML 2016 · similarity 0.88)
- Stability and Generalization of Stochastic Gradient Methods for Minimax Problems (ICML 2021 · similarity 0.88)
- Convergence of Adversarial Training in Overparametrized Neural Networks (NeurIPS 2019 · similarity 0.88)
- Implicit Bias of Gradient Descent based Adversarial Training on Separable Data (ICLR 2020 · similarity 0.88)
- Towards Understanding the Dynamics of the First-Order Adversaries (ICML 2020 · similarity 0.88)
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.