Research map / Learning theory and algorithms / Submodular and game algorithms
Optimal Continuous DR-Submodular Maximization and Applications to Provable Mean Field Inference
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 DR-DoubleGreedy instead of Submodular-DoubleGreedy | One-epoch mean-field optimization · Amazon Baby Registries ELBO and PA-ELBO experiments | objective value: The paper reports improvement in both objectives across the experiments. | Helped |
| Use DR-DoubleGreedy instead of BSCB | One-epoch mean-field optimization · Amazon Baby Registries ELBO | ELBO: 2.771±0.123 -> 3.035±0.059 | Helped |
| Use DR-DoubleGreedy instead of BSCB | One-epoch mean-field optimization · Amazon Baby Registries PA-ELBO | PA-ELBO: media, D=2: 1.477±0.128 -> 1.336±0.101; furniture, D=2: 2.287±0.399 -> 2.402±0.159 | Mixed |
| Initialize multi-epoch coordinate optimization with the DR-DoubleGreedy 1/2-guarantee solution | DG-MeanField-1/2 · Amazon Baby Registries ELBO and PA-ELBO | objective value after one epoch | Helped |
| Use random or fixed-boundary initialization rather than a guaranteed Double Greedy initializer | Multi-epoch coordinate ascent · Amazon Baby Registries ELBO and PA-ELBO | convergence behavior | Mixed |
| Run more coordinate-wise optimization epochs after initialization | Multi-epoch mean-field algorithms · Amazon Baby Registries ELBO and PA-ELBO | objective value | No clear effect |
| Use shrunken Frank-Wolfe instead of DG-MeanField-1/2 | Continuous DR-submodular optimization · Synthetic and Amazon experiments | objective value: 10 times more computations | Hurt |
| Use the gain-weighted convex combination of the two coordinate-wise candidates in the DR-DoubleGreedy update | Continuous DR-submodular maximization · Box-constrained DR-submodular objectives | approximation ratio: 1/2 | Helped |
Closest papers by meaning
- Continuous DR-submodular Maximization: Structure and Algorithms (NeurIPS 2017 · similarity 0.91)
- Provable Variational Inference for Constrained Log-Submodular Models (NeurIPS 2018 · similarity 0.90)
- Variational Inference in Mixed Probabilistic Submodular Models (NeurIPS 2016 · similarity 0.90)
- Optimization for Approximate Submodularity (NeurIPS 2018 · similarity 0.87)
- Mean Field for the Stochastic Blockmodel: Optimization Landscape and Convergence Issues (NeurIPS 2018 · similarity 0.87)
- Online Two-Stage Submodular Maximization (NeurIPS 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.