AI Paper Atlas

Research map / Learning theory and algorithms / Optimal transport

Mixed-Curvature Tree-Sliced Wasserstein Distance

Duy-Tung Pham, Viet-Hoang Tran, Thieu Vo, Tan Minh Nguyen

ICLR 2026 · Learning theory and algorithms › Optimal transport · approach: Sliced Wasserstein variants · 0 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
replace product-space or ambient slicing with MCTSW in gradient flowsliced Wasserstein baselines · Gradient flow (6 WND mixture)log W2: 0.31 -> -3.65Improved
use MCTSW as uniformity loss in a mixed-curvature latent spaceSSGE · Cora (graph SSL)accuracy: 79.55 -> 79.86Improved
use mixed-curvature latent instead of constant-curvature (E/H/S) TSWTSW-SSGE · Cora (graph SSL)accuracy: 78.33 -> 79.86Improved
replace KL regularizer with MCTSW in mixed-curvature VAEM-VAE · CIFAR-10 VAEtest BCE: 0.6419 -> 0.6000Improved
MCTSW in mixed-curvature latent vs SW-based constant-curvature variantsHSW-VAE · CIFAR-10 VAEtest BCE: 0.6012 -> 0.6000Improved
increase max curvature magnitude to 4 (from Euclidean 0) (ablation)MCTSW-SSGE · Cora (graph SSL)accuracy: 77.9 -> 79.9 -> 77.3 (maxK 0, 4, 8)Mixed
increase number of subspaces from 8 to 32 (ablation)MCTSW-SSGE · Cora (graph SSL)accuracy: 76.7 -> 79.9Improved
increase number of subspaces from 32 to 128 (ablation)MCTSW-SSGE · Cora (graph SSL)accuracy: 79.9 -> 78.7Hurt
vary number of rays per tree k from 4 to 512 (ablation)MCTSW-SSGE · Cora (graph SSL)accuracy: 79.6 (k=4), 79.9 (k=8), 79.5 (k=512)No clear effect
use spherical projection instead of orthogonal projection (with orthogonal split) (ablation)MCTSW · Gradient flow (6 WND mixture)log W2: -1.7517 -> -3.6576Improved
use orthogonal split instead of spherical split (with spherical projection) (ablation)MCTSW · Gradient flow (6 WND mixture)log W2: -3.5546 -> -3.6576; run time 1110 -> 303Improved
use simplified tree sampling (closed-form distance to ray) instead of numerical ternary search (ablation)MCTSW · Gradient flow (6 WND mixture)log W2: -3.6576 vs -3.6569; run time 303 vs 3479No 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