Research map / Optimization / Nonconvex and smooth optimization
Continuous Relaxation of MAP Inference: A Nonconvex Perspective
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 multilinear-decomposition ADMM instead of BCD, projected gradient descent, or Frank-Wolfe | Nonconvex MAP relaxation · Inpainting, feature matching, pairwise stereo, segmentation, and second-order stereo | energy | Helped |
| Use multilinear-decomposition ADMM instead of competing optimization algorithms | Nonconvex MAP relaxation · Vision MRF models | energy | Mixed |
| Use MPLP-C instead of ADMM | Feature matching · Feature matching | energy: 75.12 -> 21.22 | Helped |
| Use TRWS instead of ADMM | Pairwise stereo · Pairwise stereo | energy: 1624106.00 -> 1587961.67 | Helped |
| Use SRMP instead of ADMM | Segmentation · Segmentation | energy: 1400.68 -> 1400.57 | Helped |
| Use SRMP instead of ADMM | Second-order stereo · Second-order stereo | energy: 14068.53 -> 13914.82 | Helped |
| Use the simplex constraint set for every decomposed variable instead of looser nonnegative sets for all but one (ablation) | ADMM · MAP inference models | energy | Hurt |
| Use one simplex constraint set and nonnegative orthants for the remaining decomposed variables (ablation) | ADMM · MAP inference models | energy | Helped |
| Increase the penalty parameter when residual improvement stalls (ablation) | ADMM · MAP inference models | convergence | Helped |
| Use a small rather than a large ADMM penalty parameter (ablation) | ADMM · MAP inference models | energy | Helped |
| Use line search instead of diminishing step sizes (ablation) | Projected gradient descent and Frank-Wolfe · MAP inference models | convergence | Helped |
| Use cyclic, star, or symmetric decomposition (ablation) | ADMM · MAP inference models | energy | No clear effect |
Closest papers by meaning
- Newton-Type Methods for Inference in Higher-Order Markov Random Fields (CVPR 2017 · similarity 0.89)
- Relaxation-Based Preprocessing Techniques for Markov Random Field Inference (CVPR 2016 · similarity 0.89)
- Exact MAP-Inference by Confining Combinatorial Search With LP Relaxation (AAAI 2018 · similarity 0.88)
- Efficient semidefinite-programming-based inference for binary and multi-class MRFs (NeurIPS 2020 · similarity 0.88)
- MAP Inference via Block-Coordinate Frank-Wolfe Algorithm (CVPR 2019 · similarity 0.88)
- Continuous DR-submodular Maximization: Structure and Algorithms (NeurIPS 2017 · similarity 0.88)
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.