Research map / Learning theory and algorithms / Prediction and decision losses
Moreau-Yosida $f$-divergences
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 Moreau-Yosida regularization (β>0) to the f-divergence GAN objective | MYf-GAN · CIFAR-10 | training convergence | Helped |
| Use the Moreau-Yosida relaxation instead of the hard Lipschitz constraint (ablation) | MYf-GAN · CIFAR-10 | performance | No clear effect |
| Use the quotient critic that subtracts its output at a fixed base point (ablation) | MYf-GAN · CIFAR-10 | performance | Helped |
| Use the quotient critic instead of an unconstrained critic (ablation) | MYf-GAN · CIFAR-10 | critic output drift | Helped |
| Use the generated-sample critic score as generator loss instead of the conjugate objective (ablation) | Reverse MYf-GAN · CIFAR-10 | performance | Helped |
| Replace the generated-sample critic-score generator loss with the conjugate loss (ablation) | Reverse MYf-GAN · CIFAR-10 | performance | Hurt |
| Use the trivial divergence instead of a nontrivial f-divergence (ablation) | MYf-GAN · CIFAR-10 | FID | Hurt |
| Decay β from 0.5 to 0.2 in the α=∞ setting (ablation) | MYf-GAN · CIFAR-10 | IS | Helped |
| Increase β in the unconstrained Wasserstein variant (ablation) | MYf-GAN, α=2 · CIFAR-10 | performance: 8.14 IS; 13.33 FID | Helped |
| Use the unconstrained Wasserstein variant instead of constrained WGAN with gradient penalty | MYf-GAN, trivial, α=2, β=0.5 · CIFAR-10 | FID: 13.40 to 13.33 | Helped |
| Train the original f-GAN objective rather than the MYf-GAN regularized objective (ablation) | χ² f-GAN · CIFAR-10 | FID: 40.64 | Hurt |
Closest papers by meaning
- Bridging the Gap Between f-divergences and Bayes Hilbert Spaces (ICLR 2025 · similarity 0.89)
- Function-space regularized Rényi divergences (ICLR 2023 · similarity 0.89)
- f-Divergence Variational Inference (NeurIPS 2020 · similarity 0.86)
- $f$-Divergence Based Classification: Beyond the Use of Cross-Entropy (ICML 2024 · similarity 0.86)
- Exploring Jacobian Inexactness in Second-Order Methods for Variational Inequalities: Lower Bounds, Optimal Algorithms and Quasi-Newton Approximations (NeurIPS 2024 · similarity 0.84)
- Asymptotics of Alpha-Divergence Variational Inference Algorithms with Exponential Families (NeurIPS 2024 · similarity 0.83)
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.