Research map / Learning theory and algorithms / Optimal transport
Inverse Entropic Optimal Transport Solves Semi-supervised Learning via Data Likelihood Maximization
Mikhail Persiianov, Arip Asadulaev, Nikita Andreev, Nikita Starodubcev, Dmitry Baranchuk, Anastasis Kratsios, Evgeny Burnaev, Alexander Korotin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add unpaired samples to the paired-only likelihood objective (ablation) | EBiEOT-GMM · Weather prediction (90 paired) | test log-likelihood: 6.5 -> 20.5 | Improved |
| use EBiEOT-GMM instead of naive semi-supervised likelihood baselines | CondNF (SS) / CGMM (SS) · Weather prediction (90 paired, 500 unpaired) | test log-likelihood: 0.52 -> 20.5 (vs CondNF (SS)); 0.32 -> 20.5 (vs CGMM (SS)) | Improved |
| add unpaired data to CondNF (semi-supervised variant) (ablation) | CondNF · Gaussian to Swiss Roll | conditional distribution recovery | No clear effect |
| semi-supervised CGMM trained with the combined objective | CGMM (SS) · Gaussian to Swiss Roll | conditional distribution recovery | Hurt |
| heuristic key-point guided objectives (OTCS, FSBM) vs EBiEOT likelihood | OTCS, FSBM · Gaussian to Swiss Roll | bi-modal conditional recovery | Hurt |
| replace FSBM with EBiEOT in ALAE latent space | FSBM · Woman-to-Man translation | FID: 10.2 -> 9.3 | Improved |
| replace FSBM with EBiEOT in ALAE latent space | FSBM · Old-to-Young translation | FID: 11.5 -> 9.4 | Improved |
| replace FSBM with EBiEOT training | FSBM · Woman-to-Man translation | training time: 5 hours -> 3 minutes on A100 | Improved |
| use fully neural parameterization (EBiEOT-NN) instead of Gaussian mixture parameterization (ablation) | EBiEOT · Gaussian to Swiss Roll | training stability | Hurt |
| increase paired samples from 128 to 16k for EBiEOT-NN (ablation) | EBiEOT-NN · Gaussian to Swiss Roll | fit to target distribution | Improved |
| increase paired colored-MNIST pairs from 10 to 200 (ablation) | EBiEOT-NN · Colored MNIST hue shift | translation quality | Improved |
| deterministic-map methods (GNOT, DCPEME, parOT) on multimodal conditionals | GNOT, DCPEME, parOT · Gaussian to Swiss Roll | conditional recovery | Hurt |
Closest papers by meaning
- Generative Conditional Distributions by Neural (Entropic) Optimal Transport (ICML 2024 · similarity 0.90)
- Pairwise Optimal Transports for Training All-to-All Flow-Based Condition Transfer Model (NeurIPS 2025 · similarity 0.89)
- Sparsistency for inverse optimal transport (ICLR 2024 · similarity 0.88)
- Unbalanced minibatch Optimal Transport; applications to Domain Adaptation (ICML 2021 · similarity 0.87)
- Optimal Transport-Guided Conditional Score-Based Diffusion Model (NeurIPS 2023 · similarity 0.87)
- Provable Posterior Sampling with Denoising Oracles via Tilted Transport (NeurIPS 2024 · similarity 0.86)
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.
