CyclOT framework learns quadratic optimal transport maps from unpaired data

Researchers have introduced CyclOT, a novel neural framework for learning quadratic optimal transport maps from unpaired samples in high dimensions. This bidirectional approach utilizes synchronized forward-backward interpolants and a training objective that combines bidirectional quadratic action, discriminator-restricted Jensen-Shannon endpoint objectives, and cycle consistency. The method does not require precomputed sample pairings or explicit convex-potential parameterization. Theoretical results demonstrate its ability to recover the optimal transport maps under specific conditions, with experiments on various datasets like MNIST and CelebA validating its performance. AI

IMPACT Introduces a new method for learning optimal transport maps, potentially improving generative models and data analysis techniques.

RANK_REASON Academic paper detailing a new method for learning optimal transport maps. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• Brenier
• CelebA
• Hugging Face
• Jensen-Shannon divergence
• MNIST database
• Swiss roll

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →