Research map / Probabilistic and scientific modeling / Variational and generative sampling
Non-parametric Structured Output Networks
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 |
|---|---|---|---|
| Non-parametric structured output network (NSON) vs. parametric or unstructured posteriors on pixel stripes | NGM-100 model · Pixel stripes (COCO) test set | expected test log-likelihood: +16.46 vs +11.51 for mixture density (best baseline) | Helped |
| Factored graphical structure (chain) over non-parametric potentials vs. global mixture density | NGM-100 model · Pixel stripes (COCO) test set | expected test log-likelihood: +16.46 vs +11.51 | Helped |
| Estimating kernel locations only (no weights or bandwidths) in NGM (ablation) | NGM-100 model · Pixel stripes (COCO) test set | expected test log-likelihood: +15.26 vs +16.46 for full model | Hurt |
| Adding kernel weight estimation to bandwidth estimation (+W over -W +B) (ablation) | NGM-100 model · Pixel stripes (COCO) test set | expected test log-likelihood: +16.46 vs +15.30 | Helped |
| Recurrent inference network (RIN) vs. particle BP and black-box VI for marginal inference | RIN-100 · Pixel stripes marginal inference | marginal log-likelihood: +16.62 vs +8.85 (P-BP 400) and +3.03 (BB-VI 800) | Helped |
| Recurrent inference network vs. particle BP with 100 particles in runtime | RIN-100 · Pixel stripes marginal inference | runtime (sec): 0.04 s vs 2.11 s for P-BP-100 (over 50x faster) | Helped |
| Adding more particles to particle BP | P-BP · Pixel stripes marginal inference | marginal log-likelihood: +2.91 (50 particles) to +8.85 (400 particles) | Helped |
Closest papers by meaning
- Deep Structured Prediction with Nonlinear Output Transformations (NeurIPS 2018 · similarity 0.88)
- Modeling Structure with Undirected Neural Networks (ICML 2022 · similarity 0.87)
- On Neural Networks as Infinite Tree-Structured Probabilistic Graphical Models (NeurIPS 2024 · similarity 0.86)
- Learning Approximate Inference Networks for Structured Prediction (ICLR 2018 · similarity 0.86)
- BMRS: Bayesian Model Reduction for Structured Pruning (NeurIPS 2024 · similarity 0.86)
- SOInter: A Novel Deep Energy-Based Interpretation Method for Explaining Structured Output Models (ICLR 2024 · similarity 0.84)
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.