Research map / Learning theory and algorithms / Optimal transport
Sparsity-Constrained Optimal Transport
Tianlin Liu, Joan Puigcerver, Mathieu Blondel
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 |
|---|---|---|---|
| Use sparsity-constrained OT routing instead of S-BASE | V-MoE B/32 router · JFT-300M validation | precision@1: 44.26 -> 44.30 | Improved |
| Use sparsity-constrained OT routing instead of TopK | V-MoE B/32 router · ImageNet 10-shot | accuracy: 65.91 -> 66.52 | Improved |
| Use sparsity-constrained OT routing instead of S-BASE | V-MoE B/16 router · JFT-300M validation | precision@1: 49.80 -> 50.29 | Improved |
| Use sparsity-constrained OT routing instead of S-BASE | V-MoE B/16 router · ImageNet 10-shot | accuracy: 72.26 -> 72.76 | Improved |
| Use sparsity-constrained routing instead of TopK | V-MoE · V-MoE B/32 and B/16 training | TPUv2-core-days: 97.11 -> 122.66; 308.14 -> 433.56 | Hurt |
| Use sparsity-constrained OT instead of Soft K-Means | MNIST clustering · MNIST test | average cost: 157.10 ± 0.19 -> 157.53 ± 0.06 | Hurt |
| Use sparsity-constrained OT instead of Negentropy OT | MNIST clustering · MNIST test | KL divergence: 0.000029 ± 0.000002 -> 0.000013 ± 0.000001 | Improved |
| Increase cardinality k (ablation) | Sparsity-constrained OT solver · synthetic OT solver experiments | convergence speed | Improved |
| Use LBFGS instead of ADAM | Sparsity-constrained OT optimization · dual and semi-dual optimization | iterations to convergence | Improved |
| Use ADAM instead of LBFGS | Sparsity-constrained OT optimization · dual and semi-dual optimization | wallclock convergence speed | Mixed |
Closest papers by meaning
- The Sparse-Plus-Low-Rank Quasi-Newton Method for Entropic-Regularized Optimal Transport (ICML 2025 · similarity 0.90)
- Sparsistency for inverse optimal transport (ICLR 2024 · similarity 0.90)
- Safe and Sparse Newton Method for Entropic-Regularized Optimal Transport (NeurIPS 2024 · similarity 0.89)
- Decreasing Entropic Regularization Averaged Gradient for Semi-Discrete Optimal Transport (NeurIPS 2025 · similarity 0.88)
- Low-Rank Optimal Transport through Factor Relaxation with Latent Coupling (NeurIPS 2024 · similarity 0.88)
- Expectile Regularization for Fast and Accurate Training of Neural Optimal Transport (NeurIPS 2024 · 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.
