Normalizing flows are constructed by learning the velocity of a stochastic interpolant via a quadratic loss derived from its probability current, yielding an efficient ODE-based alternative to diffusion models.
Title resolution pending
22 Pith papers cite this work. Polarity classification is still indexing.
22
Pith papers citing it
citation-role summary
background 2
method 2
citation-polarity summary
representative citing papers
SPIDeC methods achieve arbitrarily high-order accuracy for positive dynamical systems while unconditionally preserving positivity and equilibria via a multiplicative Volterra structure, and they are L-stable with asymptotic logarithmic contractivity under Gauss-Radau nodes.
The curvature-aware precision controller adapts between FP32 and FP64 during PINN training to match double-precision accuracy at reduced computational cost.
Composed BDF schemes with complex coefficients increase order by one, supply order p+1 error estimates from the imaginary part, break the Dahlquist barrier up to order 8, and give step-ratio stability bounds for non-uniform meshes.
RedEigCD enables stable timestep increases up to 40 times larger than full-order models for projection-based ROMs of incompressible flows by using exact spectral bounds on reduced convective and diffusive operators together with a proof that ROM stable timesteps are at least as large as FOM ones.
Bi-CFM learns bidirectional mappings between initial and final state distributions to solve ill-posed inverse problems in chaotic systems, reporting metric improvements and speedups on Lorenz variants plus conservation-respecting results on three-body and globular cluster data.
Evolutionary selection on reservoir size, connectivity, spectral radius, input scaling, and regularization for Kuramoto-Sivashinsky forecasting reveals a conserved stochastic-block-model spectral envelope, locked intermediate modularity, and a horizontal cost-modularity floor in elite architectures.
INDEQS is a graph-informed NCDE variant that separates inner hidden-state mixing from outer vector-field mixing and reports lower MAE than uninformed NCDEs on synthetic advection data and real river/traffic tasks when the graph is known.
A beta-VAE analysis of pop-cosmos models finds that five latent dimensions capture the rest-frame optical SED, corresponding to stellar mass, recent star formation, dust, and two gas ionization states.
PIDM-DP integrates Dormand-Prince ODE solving into DDPM denoising with scheduled physics guidance to reconstruct chaotic states, reporting up to 15.4x RMSE gains over baselines on five systems including stiff cases.
SITN performs single-sample OOD detection via goodness-of-fit testing on noise samples in the factorised latent space of continuous normalizing flows.
A simulation-based inference method with Gaussian process emulators trained on 1300 kilonova simulations recovers parameters accurately and rapidly while avoiding MCMC biases from likelihood misspecification.
Nowhere-vanishing Koopman eigenfunctions form a multiplicative group, enabling polynomial extensions from principal ones to enrich eigenspaces and enable global representations from local data in multistable systems.
Galaxy size-mass relations exhibit double power-law breaks at different pivot masses for quiescent versus bulge-dominated samples, coinciding with AGN activity scales.
Flow matching achieves single-step pixel accuracy and 20-step perceptual quality for Sentinel-2 super-resolution, outperforming diffusion and Real-ESRGAN while enabling large-scale 2.5 m land-cover products.
Polytropic stellar wind models are extended beyond extreme adiabatic cases to non-adiabatic localized heating, with added energy shown plausible relative to flares and relevant to solar wind observations.
BG-SINDy reformulates l0-constrained regression as term-level l2,0 regularization and uses progressive pruning guided by balance contributions to recover small-coefficient terms in multiscale PDEs.
Bayesian-ARGOS is a hybrid frequentist-Bayesian method that discovers equations from limited noisy observations more efficiently than SINDy or bootstrap-ARGOS while adding uncertainty quantification.
S-MNN reformulates Mechanistic Neural Networks to achieve linear computational complexity for long sequences while preserving accuracy and interpretability.
The Parker-Sochacki method delivers 4 to 13 orders of magnitude better kinetic energy conservation than Runge-Kutta methods for charged particle motion in static magnetic fields while running faster at matched accuracy.
Comparative experiments on three chaotic systems find that architectures using integrator-like updates exhibit lower bias, reduced perturbation amplification, and more stable long-horizon rollouts than other common designs when model capacity is matched.
Experimental imbibition data on four porous materials are pre-processed with a monotonicity-preserving fit and used to calibrate a PDE model of capillary absorption.
citing papers explorer
No citing papers match the current filters.