Research map / Optimization / Optimizers
Principled Parallel Mean-Field Inference for Discrete Random Fields
Check 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 proximal-gradient parallel mean-field updates with KL damping in natural-parameter space instead of ADHOC damping | DBN · DBN (1.00s) | KL divergence: -2047.31 -> -2130.61 | Helped |
| Use proximal-gradient parallel mean-field updates with KL damping in natural-parameter space instead of ADHOC damping | GRID · GRID (1.00s) | KL divergence: -18349.03 -> -18252.48 | Helped |
| Use proximal-gradient parallel mean-field updates with KL damping in natural-parameter space instead of ADHOC damping | People Detection · People Detection (5s) | MODA: 0.829 -> 0.829 | No clear effect |
| Use proximal-gradient parallel mean-field updates with KL damping in natural-parameter space instead of ADHOC damping | Semantic Segmentation with co-occurrence potentials · Semantic Segmentation (50s) | I/U: 68.04 -> 68.44 | Helped |
| Add adaptive anisotropic damping to fixed natural-parameter damping (ablation) | People Detection · People Detection (5s) | MODA: 0.829 -> 0.829 | No clear effect |
| Add momentum to adaptive proximal mean-field updates (ablation) | People Detection · People Detection (5s) | MODA: 0.829 -> 0.826 | Hurt |
| Add ADAM-style adaptive updates to adaptive proximal mean-field updates (ablation) | People Detection · People Detection (5s) | MODA: 0.829 -> 0.824 | Hurt |
| Add momentum to fixed natural-parameter updates (ablation) | DBN · DBN (1.00s) | KL divergence: -2081.91 -> -2374.51 | Helped |
| Add ADAM-style adaptive updates to fixed natural-parameter updates (ablation) | GRID · GRID (1.00s) | KL divergence: -18219.45 -> -18740.36 | Helped |
| Add ADAM-style adaptive updates to fixed natural-parameter updates (ablation) | Semantic Segmentation · Semantic Segmentation (50s) | I/U: 68.44 -> 68.47 | Helped |
| Add ADAM-style adaptive updates to fixed natural-parameter updates (ablation) | Characters Inpainting · Characters Inpainting (3s) | pixel accuracy (PA): 62.35 -> 67.12 | Helped |
| Add co-occurrence potentials and use convergent proximal inference rather than the original model's inference | DeepLab-CRF · Semantic Segmentation (50s) | I/U: 68.00 -> 68.44 | Helped |
Closest papers by meaning
- Multi-Modal Mean-Fields via Cardinality-Based Clamping (CVPR 2017 · similarity 0.89)
- Mean Field for the Stochastic Blockmodel: Optimization Landscape and Convergence Issues (NeurIPS 2018 · similarity 0.88)
- Fast Black-box Variational Inference through Stochastic Trust-Region Optimization (NeurIPS 2017 · similarity 0.87)
- Marginalized Stochastic Natural Gradients for Black-Box Variational Inference (ICML 2021 · similarity 0.87)
- Nearly Dimension-Independent Convergence of Mean-Field Black-Box Variational Inference (NeurIPS 2025 · similarity 0.87)
- Robust, Accurate Stochastic Optimization for Variational Inference (NeurIPS 2020 · similarity 0.87)
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.