optimal-transport

This paper proposes a scalable local Sinkhorn divergence framework for training stochastic neural networks to reconstruct multidimensional random fields, with theoretical generalization error bounds and numerical demonstrations for uncertainty quantification.

This paper proposes PairAlign, a pair-centric graph rewiring framework that uses optimal transport-guided communication alignment to alleviate over-squashing in message-passing neural networks, with theoretical analysis and experiments on standard benchmarks.

This paper introduces PMOT, a potential-flow framework for general p-cost optimal transport using continuous normalizing flows, with theoretical zero-loss exactness and promising results on synthetic and high-dimensional benchmarks.

This paper proposes a unified optimal transport framework for cold-start active learning, introducing a Sinkhorn-based algorithm (ε-AS) that adapts regularization strength to data and achieves state-of-the-art results on six datasets, including improving ImageNet-1k accuracy by 1.29% over prior methods while reducing selection time by 56.2%.

Proposes Quantile Coupling Flow Matching (QC-FM), a lightweight one-sided coupling that constructs source samples from data ranks along random directions without needing pairwise cost matrices or assignment. Achieves up to 12.9% FID improvement over baseline on CIFAR-10, CelebA, FFHQ, and ImageNet-64.

DualAnchor is a training framework for gloss-free sign language translation that uses token-level prior anchoring to preserve LLM language priors and optimal transport alignment to improve lexical fidelity, achieving strong results on PHOENIX-2014T and CSL-Daily.

This paper proposes formulating the bridge between planned tasks and performed actions as a quasi-linear Fisher market, allowing fractional credit assignment. It introduces instruments for conservation and junk filtering, and extends the model with entropy regularization to handle noise, unifying it with optimal transport.

This paper introduces TAKE (Trajectory-Aware Knowledge Estimation), a text dataset distillation framework that uses influence functions and optimal transport to reduce datasets to as little as 0.1% of their original size while preserving downstream task fidelity.

FastAlign presents a scalable, sparsity-aware framework for optimal transport-based network alignment, achieving state-of-the-art accuracy while reducing runtime by up to 9.45x on CPU and 32.54x on GPU.

VTaMo introduces explicit multi-granularity video-text alignment for sign language translation using optimal transport and contrastive learning, achieving state-of-the-art performance on four benchmarks.

This paper introduces Reward Transport, a method that uses optimal transport coupling during flow matching training to align a scalar noise coordinate with molecular rewards, enabling monotone control over molecular properties like logP and QED at inference without additional computation.

Introduces Diffusion Semi-Relaxed Fused Gromov-Wasserstein (DsrFGW), a novel method that integrates node features and structural connectivity for graph comparison via optimal transport and diffusion processes, demonstrating improved robustness to noise and missing edges on synthetic tasks.

Proposes SAOT, a structure-aware optimal transport framework for self-supervised continual graph learning that preserves relational structure across tasks. Achieves significant performance gains over state-of-the-art methods on multiple benchmarks, including up to 15% improvement on Products-CL.

OTCache is a training-free framework that uses optimal transport to predict caching schedules for diffusion models, achieving up to 4.7x acceleration on FLUX.1, Qwen-Image, and HunyuanVideo while improving generation fidelity.

The paper introduces ReMatch, a method that aligns training residual distributions to test-time regimes via optimal transport in PCA space to mitigate bias in probabilistic downscaling, achieving better calibration and dispersion.

This paper reveals that diffusion models and flow matching are two sides of the same Wasserstein geometry: diffusion follows a free-energy gradient flow (initial-value problem), while flow matching follows a Wasserstein geodesic (boundary-value problem), and they are unified through the JKO scheme.

MeshFlow introduces an equivariant optimal-transport flow matching model for direct triangle mesh generation, achieving state-of-the-art quality while providing approximately 18x inference speedup over autoregressive methods.

This paper proposes RicciBind, a geometric representation framework that integrates Ricci curvature and optimal transport for protein-ligand binding affinity prediction, demonstrating superior accuracy and interpretability across benchmarks.

This paper extends optimal transport-based hallucination detection to all decoder layers in NMT and abstractive summarization, finding that detection is concentrated in early layers and that the geometric signal transfers poorly to summarization due to faithfulness failures not detectable via attention concentration.

This paper extends MST-Direct, a method for multivariate geostatistical simulation using Sinkhorn optimal transport, from bivariate unconditional small-grid settings to multivariate, conditional, and large-scale settings, preserving joint distributions exactly and outperforming existing methods.