Research map / Learning theory and algorithms / Optimal transport
Solving General Elliptical Mixture Models through an Approximate Wasserstein Manifold
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 |
|---|---|---|---|
| Optimize GMMs with the approximate Wasserstein manifold and Dadam instead of WO/M | GMM estimation · Synthetic GMM, m=16, k=27 | Wass: 5e3 ± 589 -> 2.39 ± 0.33 | Helped |
| Optimize GMMs with the approximate Wasserstein manifold and Dadam instead of WO/M | GMM estimation · Synthetic GMM, m=16, k=27 | NLL: Inf -> 44.92 ± 1.04 | Helped |
| Optimize GMMs with the approximate Wasserstein manifold and Dadam instead of EM | GMM estimation · Synthetic GMM, m=16, k=27 | Wass: 2.51 ± 1.27 -> 2.39 ± 0.33 | Helped |
| Optimize GMMs with the approximate Wasserstein manifold and Dadam instead of EM | GMM estimation · Synthetic GMM, m=16, k=27 | NLL: 48.84 ± 1.24 -> 44.92 ± 1.04 | Helped |
| Use the approximate Wasserstein manifold and Dadam instead of WO/M | Gaussian mixture · BSDS500 | Wass: 89.5 ± 10.0 -> 35.0 ± 4.68 | Helped |
| Use the approximate Wasserstein manifold and Dadam instead of IRA | Gaussian mixture · BSDS500 | Wass: 53.6 ± 21.9 -> 35.0 ± 4.68 | Helped |
| Use the approximate Wasserstein manifold and Dadam instead of IRA | Logistic mixture · BSDS500 | Wass: 68.0 ± 24.2 -> 45.8 ± 12.7 | Helped |
| Use the approximate Wasserstein manifold and Dadam instead of IRA | Cauchy mixture · BSDS500 | NLL: 12.0 ± 0.05 -> 12.1 ± 0.01 | Hurt |
| Use Dadam rather than vanilla manifold gradient descent (ablation) | Elliptical mixture covariance optimization · Synthetic data | convergence speed | Helped |
| Use Dadam rather than Riemannian Adam (ablation) | Elliptical mixture covariance optimization · Synthetic data | convergence speed | Helped |
| Optimize on the approximate Wasserstein manifold rather than without a manifold (ablation) | Elliptical mixture estimation · Synthetic data | convergence speed | Helped |
| Use Dadam for covariance updates rather than element-wise adaptive second-moment accumulation (ablation) | Elliptical mixture covariance optimization · Synthetic data | convergence speed | Helped |
Closest papers by meaning
- Refined Convergence Rates for Maximum Likelihood Estimation under Finite Mixture Models (ICML 2022 · similarity 0.91)
- Sliced Wasserstein Distance for Learning Gaussian Mixture Models (CVPR 2018 · similarity 0.89)
- Toward Global Convergence of Gradient EM for Over-Paramterized Gaussian Mixture Models (NeurIPS 2024 · similarity 0.87)
- Sublinear Variational Optimization of Gaussian Mixture Models with Millions to Billions of Parameters (NeurIPS 2026 · similarity 0.85)
- Is $\sqrt{d}$ Separation Necessary for Gradient EM to Learn Gaussian Mixtures in High Dimensions? (NeurIPS 2026 · similarity 0.85)
- Big Learning Expectation Maximization (AAAI 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.