Research map / Probabilistic and scientific modeling / Variational and generative sampling
Boosted Generative Models
Aditya Grover, Stefano Ermon
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 |
|---|---|---|---|
| Replace additive boosting with multiplicative discriminative boosting | Mixture-of-two-Gaussians base model · Mixture of Gaussians test set | NLL (nats): 4.69 ± 0.01 -> 4.35 ± 0.01 | Improved |
| Use negative cross-entropy instead of Hellinger distance for the discriminator (ablation) | DiscBGM · Mixture of Gaussians test set | NLL (nats): 4.35 ± 0.01 -> 4.42 ± 0.01 | Hurt |
| Use additive rather than multiplicative boosting (ablation) | Generative boosting · Mixture of Gaussians test set | NLL (nats): 4.58 ± 0.10 -> 4.64 ± 0.02 | Hurt |
| Add discriminative multiplicative boosting | Mixture of Bernoullis · Density estimation across benchmark datasets | NLL (nats): 42.23 -> 34.51; 11.27 -> 10.91; 55.67 -> 34.93; 99.42 -> 98.45; 11.72 -> 11.13; 63.13 -> 54.79 | Improved |
| Add discriminative multiplicative boosting | Sum-Product Network · Density estimation across benchmark datasets | NLL (nats): 31.08 -> 28.09; 14.94 -> 10.88; 26.70 -> 23.69; 92.60 -> 86.63; 12.71 -> 10.67; 19.19 -> 17.82 | Improved |
| Add discriminative multiplicative boosting | Mixture of Bernoullis · Classification across benchmark datasets | prediction accuracy: 0.8395 -> 0.9043; 0.9776 -> 0.9792; 0.8461 -> 0.9267; 0.7517 -> 0.7526; 0.9817 -> 0.9831; 0.9922 -> 0.9927 | Improved |
| Add discriminative multiplicative boosting | Sum-Product Network · Classification across benchmark datasets | prediction accuracy: 0.9258 -> 0.9416; 0.9780 -> 0.9791; 0.9599 -> 0.9636; 0.7799 -> 0.7811; 0.9824 -> 0.9838; 0.9982 -> 0.9982 | Mixed |
| Use additive boosting rather than the base model (ablation) | Mixture of Bernoullis · Density estimation across benchmark datasets | NLL (nats): 42.23 -> 43.19; 11.27 -> 12.24; 55.67 -> 55.91; 99.42 -> 100.37; 11.72 -> 12.57; 63.13 -> 63.73 | Hurt |
| Replace the base model with GenBGM (ablation) | Mixture of Bernoullis · Density estimation across benchmark datasets | NLL (nats): 42.23 -> 41.43; 11.27 -> 11.20; 55.67 -> 50.66; 99.42 -> 99.23; 11.72 -> 12.41; 63.13 -> 63.19 | Mixed |
| Use the uniform model-weighting heuristic (ablation) | DiscBGM · Mixture of Gaussians | NLL (nats) | Hurt |
| Use the unity model-weighting heuristic (ablation) | GenBGM · Mixture of Gaussians | NLL (nats) | Hurt |
| Use the decay model-weighting heuristic (ablation) | GenBGM and DiscBGM · Mixture of Gaussians | NLL (nats) | Improved |
Closest papers by meaning
- Generative Forests (NeurIPS 2024 · similarity 0.86)
- Boost-and-Skip: A Simple Guidance-Free Diffusion for Minority Generation (ICML 2025 · similarity 0.86)
- NRGBoost: Energy-Based Generative Boosted Trees (ICLR 2025 · similarity 0.86)
- Generative Marginalization Models (ICML 2024 · similarity 0.85)
- DiverGen: Improving Instance Segmentation by Learning Wider Data Distribution with More Diverse Generative Data (CVPR 2024 · similarity 0.85)
- Free Lunch in the Forest: Functionally-Identical Pruning of Boosted Tree Ensembles (AAAI 2025 · similarity 0.85)
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.
