arXiv:2603.10718v2 Announce Type: replace Abstract: Flow Matching enables simulation-free training of generative models on Riemannian manifolds, yet sampling typically still relies on numerically integrating a probability-flow ODE. We propose Riemannian MeanFlow (RMF), extending MeanFlow to manifold-valued generation where velocities lie in location-dependent tangent spaces. RMF defines an average-velocity field via parallel transport and derives a Riemannian MeanFlow identity that links average and instantaneous velocities for intrinsic supervision. We make this identity practical in a log-ma
The continuous evolution of generative AI research drives new methods for improving model training and sampling efficiency.
Advanced techniques like Riemannian MeanFlow can lead to more robust and powerful generative AI models, impacting various applications from data synthesis to complex simulations.
This research introduces a novel, more efficient approach to generative modeling on complex data structures (manifolds), potentially simplifying and speeding up model development.
- · AI researchers
- · Generative AI developers
- · Machine learning startups
- · Inefficient generative model architectures
Improved performance and broader applicability of generative AI models in specialized domains.
Acceleration of research into AI applications requiring nuanced handling of complex data geometries.
Potentially enables new forms of data synthesis or simulation that were previously computationally prohibitive.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG