Research map / Probabilistic and scientific modeling / Variational and generative sampling
Do Deep Generative Models Know What They Don't Know?
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 |
|---|---|---|---|
| Use likelihood as the sole indicator of whether an input matches the training distribution | Glow trained on CIFAR-10 · CIFAR-10 vs SVHN | log-likelihood | Hurt |
| Use likelihood as the sole indicator of whether an input matches the training distribution | Glow trained on FashionMNIST · FashionMNIST vs MNIST | BPD | Hurt |
| Use likelihood as the sole indicator of whether an input matches the training distribution | Glow · CelebA vs SVHN; ImageNet vs CIFAR-10, CIFAR-100, and SVHN | log-likelihood | Hurt |
| Use likelihood as the sole indicator of whether an input matches the training distribution | PixelCNN and VAE · FashionMNIST vs MNIST; CIFAR-10 vs SVHN | log-likelihood | Hurt |
| Replace affine coupling with translation-only coupling to make flow volume constant (ablation) | Glow trained on CIFAR-10 · CIFAR-10 vs SVHN | log-likelihood | No clear effect |
| Average likelihoods from five independently initialized Glow models (ablation) | Glow trained on CIFAR-10 · CIFAR-10 vs SVHN | log-likelihood | No clear effect |
| Stop training early (ablation) | NVP-Glow · CIFAR-10 vs SVHN | BPD | No clear effect |
| Reduce image variance by graying the images (ablation) | Glow · CIFAR-10 and SVHN | log-likelihood | Helped |
| Use sigmoid rather than exponential scale parametrization (ablation) | RNVP flow · Out-of-distribution detection | log-likelihood | No clear effect |
Closest papers by meaning
- Revisiting flow generative models for Out-of-distribution detection (ICLR 2022 · similarity 0.89)
- Adaptive Density Estimation for Generative Models (NeurIPS 2019 · similarity 0.89)
- Understanding Likelihood of Normalizing Flow and Image Complexity through the Lens of Out-of-Distribution Detection (AAAI 2024 · similarity 0.88)
- Understanding Failures in Out-of-Distribution Detection with Deep Generative Models (ICML 2021 · similarity 0.88)
- Flow-GAN: Combining Maximum Likelihood and Adversarial Learning in Generative Models (AAAI 2018 · similarity 0.87)
- On the Closed-Form of Flow Matching: Generalization Does Not Arise from Target Stochasticity (NeurIPS 2025 · 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.