The paper introduces Difficulty-Calibrated Flow Matching, a method that adapts the noise-to-data interpolation schedule in Conditional Flow Matching based on a pilot run’s loss profile. By setting the schedule to the quantile function of this difficulty profile, the training trajectory spends more time where the velocity is hardest to learn. Experiments on CIFAR-10, MNIST, and Fashion‑MNIST show that this calibrated path achieves the best FID on CIFAR‑10 and outperforms all fixed schedules in large‑batch, few‑update settings, where compute is most limited.
Flow matching trains a neural network to regress the conditional velocity along a linear interpolant between noise and data, and the number of network evaluations~(NFE) sets the cost of sampling. The straight-line interpolant carries an implicit choice: the sample moves at constant speed throughout the trajectory.
arXiv:2607. 11442v1 Announce Type: new Abstract: Flow matching trains a neural network to regress the conditional velocity along a linear interpolant between noise and data, and the number of network evaluations~(NFE) sets the cost of sampling.
arXiv:2605.10302v4 Announce Type: replace Abstract: Existing approaches to controllable generation typically rely on fine-tuning, auxiliary networks, or test-time search. We show that flow matching a...
I‑SplineFlow introduces a new way to learn monotone spline stochastic interpolant schedulers for few‑step generation with pretrained diffusion and flow models. By parameterizing the scheduler with integrated monotone splines (I‑splines), the method decouples polynomial degree from the number of mixture weights, enabling compact support, better‑conditioned Jacobians, and strictly monotone signal‑to‑noise ratios without ordering constraints. Experiments on EDM, ReFlow, and Simple ReFlow show that I‑SplineFlow consistently improves few‑step FID over Bézier scheduling, especially at low NFEs, while training in only minutes.
arXiv:2608. 19540v1 Announce Type: new Abstract: Training fast generators on new domains with limited data remains challenging for two reasons.
arXiv:2607. 28864v1 Announce Type: cross Abstract: Tree-based diffusion models fit flexible conditional predictive distributions for tabular regression without a neural density estimator, but they inherit their design defaults---noising path, parameterization, training distribution, features, sampler---from the neural setting.
arXiv:2605. 00941v4 Announce Type: replace Abstract: Flow matching has become a leading framework for generative modeling, but quantifying the uncertainty of its samples remains an open problem.
The paper introduces data‑space iteration, a few‑step generation framework that eliminates the need for a fixed probability‑flow discretization. By refining predictions directly in data space with a shared generator, each iteration is trained to produce the best possible sample within its capacity. Experiments on class‑conditional ImageNet 256×256 show that data‑space iteration outperforms standard discretization baselines and matches or improves upon schedule‑searched variants without schedule‑specific training.
The paper introduces model‑aware schedules for diffusion and flow‑matching generative models by leveraging fiberwise optimal transport. It defines a fiberwise prediction risk that averages optimal‑transport costs between true and predicted signal/noise decompositions, and combines this with kinetic action to derive a closed‑form optimal time allocation. Experiments on DDPMs and flow matching across various targets, datasets, and architectures show that these schedules consistently outperform strong baselines, achieving significant FID reductions and demonstrating empirical universality after normalizing risk profiles.
arXiv:2610.08717v1 Announce Type: cross Abstract: We study path-flow alignment as a unified training objective for flow matching. Instead of fixing the interpolation path and learning only the veloci...
arXiv:2610.01408v1 Announce Type: new Abstract: Trajectory crossing remains a critical bottleneck in Flow Matching (FM), and previous works typically view these crossings from a theoretical optimizat...
We study path-flow alignment as a unified training objective for flow matching. Instead of fixing the interpolation path and learning only the velocity field, we jointly train an endpoint-preserving p...