Research map / Probabilistic and scientific modeling / Variational and generative sampling
Deep Density Destructors
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 deep Gaussian copula destructor instead of MAF (5) | density model · MNIST | test log likelihood: -1300 -> -1028 | Helped |
| use image pairs (tree) destructor instead of MAF MoG | density model · MNIST | test log likelihood: -1100 -> -1003 | Helped |
| use deep copula instead of Gaussian baseline | density model · CIFAR-10 | test log likelihood: 2367 -> 2626 | Helped |
| use deep copula instead of MAF (10) | density model · CIFAR-10 | test log likelihood: 3054 -> 2626 | Hurt |
| use image pairs destructors (local pixel pairs) instead of deep copula | density model · CIFAR-10 | test log likelihood: 2626 -> -2404 | Hurt |
| use pair destructor with density tree instead of Gaussian copula (ablation) | image pairs destructor · MNIST | test log likelihood: -1043 -> -1003 | Helped |
| use deep models instead of shallow single-layer models | tree and Gaussian mixture destructors · concentric circles | test likelihood | Helped |
| increase regularization strength per layer (ablation) | random tree destructor · concentric circles | number of selected layers: 172 layers at alpha=100 | Helped |
| use density tree instead of random tree | tree destructors · concentric circles | model compactness: 53 layers vs 172 layers | Mixed |
| use more layers with weaker regularization (w=0.9, 36 layers) instead of w=0.5 with 16 layers (ablation) | Gaussian mixture destructor · concentric circles | test likelihood: 36 layers -> 16 layers | Hurt |
| compare CPU (10 CPUs) deep copula to GPU MAF | density destructors · MNIST | train time: 0.2 vs 0.1 | Mixed |
| add successive layers to deep copula (ablation) | deep Gaussian copula · CIFAR-10 | layer log likelihood: -4572 -> 2461 | Helped |
Closest papers by meaning
- GaussianUDF: Inferring Unsigned Distance Functions through 3D Gaussian Splatting (CVPR 2025 · similarity 0.83)
- DropGaussian: Structural Regularization for Sparse-view Gaussian Splatting (CVPR 2025 · similarity 0.83)
- GaussHDR: High Dynamic Range Gaussian Splatting via Learning Unified 3D and 2D Local Tone Mapping (CVPR 2025 · similarity 0.83)
- LeanGaussian: Breaking Pixel or Point Cloud Correspondence in Modeling 3D Gaussians (CVPR 2025 · similarity 0.83)
- MaskGaussian: Adaptive 3D Gaussian Representation from Probabilistic Masks (CVPR 2025 · similarity 0.83)
- MonoSplat: Generalizable 3D Gaussian Splatting from Monocular Depth Foundation Models (CVPR 2025 · 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.