AI Paper Atlas

Research map / Learning theory and algorithms / Optimal transport

Bringing regularized optimal transport to lightspeed: a splitting method adapted for GPUs

Jacob Lindbäck, Zesen Wang, Mikael Johansson

NeurIPS 2023 · Learning theory and algorithms › Optimal transport · approach: Sinkhorn and entropic OT algorithms · 1 citations (OpenAlex)

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 changedWhereMeasuredEffect
Use RDROT (Douglas-Rachford, GPU) instead of L-BFGS dual ascent for quadratic OTQDROT · Quadratic-regularized OT, 1000x1000 and 2000x3000, 50 datasetsruntime to 1e-4 accuracyImproved
Use GLDROT instead of Sinkhorn-based group-lasso OT (Ent 1e-3)GLSK · Domain adaptation simulationmedian runtime (s): 3.77 -> 0.0619Improved
Use GLDROT instead of GLSK (Ent 1e-3) for adaptation qualityGLSK · Domain adaptation simulationmedian aggregated W2 distance: 0.311 -> 0.0576Improved
Increase entropic regularization in GLSK from 1e-3 to 1e-1 (ablation)GLSK · Domain adaptation simulationmedian aggregated W2 distance: 0.311 -> 8.24Hurt
Increase entropic regularization in GLSK from 1e-1 to 1e+2 (ablation)GLSK · Domain adaptation simulationmedian runtime (s): 3.73 -> 1.86Improved
Increase entropic regularization in GLSK from 1e-1 to 1e+2 (ablation)GLSK · Domain adaptation simulationmedian aggregated W2 distance: 8.24 -> 52.8Hurt
Increase group-lasso weight in GLDROT from 1e-3 to 5e-3 (ablation)GLDROT · Domain adaptation simulationmedian runtime (s): 0.0619 -> 0.0384Improved
Increase group-lasso weight in GLDROT from 1e-3 to 5e-3 (ablation)GLDROT · Domain adaptation simulationmedian aggregated W2 distance: 0.0576 -> 0.0879Hurt
Reduce Sinkhorn entropic regularization to limit shrinkage (ablation)GLSK · Domain adaptationruntime and stabilityMixed
Use GPU RDROT with low regularization for Minibatch Energy Distance GAN lossRDROT · MNIST/CIFAR10 GAN trainingtraining timeNo clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy