Research map / Learning theory and algorithms / Sample complexity
Parameter Learning for Log-supermodular Distributions
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 perturb-and-MAP logistic bound instead of L-field bound for maximum likelihood learning | log-supermodular model parameter learning · Gaussian mixture graph-cut bound comparison | upper bound on log-partition function | Helped |
| Replace logistic bound with L-field bound in maximum likelihood | parameter learning (alpha, t) · theoretical analysis | learned alpha | Hurt |
| Use mean-marginals instead of max-marginal decoding (ablation) | supervised denoising with logistic model · Weizmann horse supervised denoising | proportion of correct pixels error: 0.0 to 0.1 points lower error at 10% and 20% noise, equal at 1% and 5% | Mixed |
| Supervised logistic model vs structured-output SVM | supervised image denoising · Weizmann horse supervised denoising | proportion of correct pixels error: error 0.4% vs 0.6% at 1% noise up to 4.1% vs 6.0% at 20% noise | Helped |
| Learn noise level pi jointly instead of fixing it in unsupervised denoising (ablation) | unsupervised logistic model with max-marginals · Weizmann horse unsupervised denoising | proportion of correct pixels error: at 5% noise error rises from 0.9% to 3.5%; at 20% from 5.3% to 20.0% | Hurt |
Closest papers by meaning
- Differentiable Learning of Submodular Models (NeurIPS 2017 · similarity 0.89)
- Provable Variational Inference for Constrained Log-Submodular Models (NeurIPS 2018 · similarity 0.89)
- BGDB: Bernoulli-Gaussian Decision Block with Improved Denoising Diffusion Probabilistic Models (AAAI 2025 · similarity 0.85)
- Sampling Binary Data by Denoising through Score Functions (ICML 2025 · similarity 0.85)
- Learning normalized image densities via dual score matching (NeurIPS 2025 · similarity 0.84)
- Sublinear Variational Optimization of Gaussian Mixture Models with Millions to Billions of Parameters (NeurIPS 2026 · 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.