Building Normalizing Flows with Stochastic Interpolants
2022/09/30 by Michael S. Albergo, Eric Vanden-Eijnden, Eric Vanden‐Eijnden +2 · 1 voice · 191 citations
Computer Science · Medicine · Physics and Astronomy · #Advanced Neuroimaging Techniques and Applications #Generative Adversarial Networks and Image Synthesis #Model Reduction and Neural Networks #cs.LG #stat.ML
paper · pdf · doi:10.48550/arxiv.2209.15571
openalex publication_date 2022/09/30 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Abstract
A generative model based on a continuous-time normalizing flow between any pair of base and target probability densities is proposed. The velocity field of this flow is inferred from the probability current of a time-dependent density that interpolates between the base and the target in finite time. Unlike conventional normalizing flow inference methods based the maximum likelihood principle, which require costly backpropagation through ODE solvers, our interpolant approach leads to a simple quadratic loss for the velocity itself which is expressed in terms of expectations that are readily amenable to empirical estimation. The flow can be used to generate samples from either the base or target, and to estimate the likelihood at any time along the interpolant. In addition, the flow can be optimized to minimize the path length of the interpolant density, thereby paving the way for building optimal transport maps. In situations where the base is a Gaussian density, we also show that the velocity of our normalizing flow can also be used to construct a diffusion model to sample the target as well as estimate its score. However, our approach shows that we can bypass this diffusion completely and work at the level of the probability flow with greater simplicity, opening an avenue for methods based solely on ordinary differential equations as an alternative to those based on stochastic differential equations. Benchmarking on density estimation tasks illustrates that the learned flow can match and surpass conventional continuous flows at a fraction of the cost, and compares well with diffusions on image generation on CIFAR-10 and ImageNet 32×32. The method scales ab-initio ODE flows to previously unreachable image resolutions, demonstrated up to 128×128.
Cited by
- An Introduction to Bayesian and Frequentist Simulation-Based Inference with Machine Learning
- FES-FM: Free Energy Surface Sampling via Reduced Flow Matching
- Boltzmann-Expected Molecular Design with Decoupled Annealing Flows
- JAGG: Jacobian-Aggregated Group Gradient for Efficient GRPO Training of Diffusion Models
- DeforM: Reasoning-Guided Physics-Aware Video Generation via Spatial-Temporal Masking
- ROMS-IMLE: A Minimalist Approach to Competitive Single-Step Generative Modelling
- Aligned Stable Inpainting: Mitigating Unwanted Object Insertion and Preserving Color Consistency
- Signed Rectified Flow: Negativity-Controlled Generation
- REGLUE Your Latents with Global and Local Semantics for Entangled Diffusion
- Curvature-Adaptive Consistency Flow Matching: Autonomous Trajectory Optimization via Reinforcement Learning
- TVGL-CFM:Generating and Forecasting Time-Varying Trajectories of Dynamic Networks with Conditional Flow Matching
- Spatial Transport of Integration Error in Generative ODEs
- Optimal Self-Distillation for Rectified Flow via Linear Probing
- Conditioning Gaussian Processes on Almost Anything
- Stable Audio 3
- Flow Matching from Viewpoint of Proximal Operators
- The Generation Phases of Flow Matching: a Denoising Perspective
- Multimodal Datasets with Controllable Mutual Information
- SimpleFold: Folding Proteins is Simpler than You Think
- On the Closed-Form of Flow Matching: Generalization Does Not Arise from Target Stochasticity
- ProtComposer: Compositional Protein Structure Generation with 3D Ellipsoids
- On Conditional Stochastic Interpolation for Generative Nonlinear Sufficient Dimension Reduction
- Fast Organic Crystal Structure Prediction with Unit Cell Flow Matching
- Probing the Geometry of Diffusion Models with the String Method
- Self-Evaluation Unlocks Any-Step Text-to-Image Generation
- Generative Modeling by Value-Driven Transport
- Escaping the Euclidean Void: Manifold-Informed Flow Matching for Sequential Recommendation
- Amortized Posteriors for Estimation of Material Constitutive Parameters from Multimodal Measurements on Small Punch Tests
- A Dacorogna-Moser construction of transport maps on ℝd with application to geodesics on the space of couplings
- FMOPF: Latent Flow Matching with Constraint-Aware Interaction Priors for AC Optimal Power Flow
- Tilt Matching for Scalable Sampling and Fine-Tuning
- Generative modeling of conditional probability distributions on the level-sets of collective variables
- FlowDet: Unifying Object Detection and Generative Transport Flows
- SoFlow: Solution Flow Models for One-Step Generative Modeling
- Expand and Prune: Maximizing Trajectory Diversity for Effective GRPO in Generative Models
- Exploring the Design Space of Transition Matching
- Bidirectional Normalizing Flow: From Data to Noise and Back
- Generative Modeling from Black-box Corruptions via Self-Consistent Stochastic Interpolants
- FALCON: Few-step Accurate Likelihoods for Continuous Flows
- ClearPotential: Revealing Local Dark Matter in Three Dimensions
- Temporal Concept Dynamics in Diffusion Models via Prompt-Conditioned Interventions
- TreeGRPO: Tree-Advantage GRPO for Online RL Post-Training of Diffusion Models
- Scalable Offline Model-Based RL with Action Chunks
- Pathway to O(√(d)) Complexity bound under Wasserstein metric of flow-based models
- NEAT: Neighborhood-Guided, Efficient, Autoregressive Set Transformer for 3D Molecular Generation
- StreamFlow: Theory, Algorithm, and Implementation for High-Efficiency Rectified Flow Generation
- COOPER: A Unified Model for Cooperative Perception and Reasoning in Spatial Intelligence
- LeMat-GenBench: A Unified Evaluation Framework for Crystal Generative Models
- UniTS: Unified Time Series Generative Model for Remote Sensing
- ReflexFlow: Rethinking Learning Objective for Exposure Bias Alleviation in Flow Matching
- Plug-and-Play Image Restoration with Flow Matching: A Continuous Viewpoint
- Glance: Accelerating Diffusion Models with 1 Sample
- From Navigation to Refinement: Revealing the Two-Stage Nature of Flow-based Diffusion Models through Oracle Velocity
- Joint Distillation for Fast Likelihood Evaluation and Sampling in Flow-based Models
- Unrolled Networks are Conditional Probability Flows in MRI Reconstruction
- Improved Mean Flows: On the Challenges of Fastforward Generative Models
- Much Ado About Noising: Dispelling the Myths of Generative Robotic Control
- Reversible Inversion for Training-Free Exemplar-guided Image Editing
- High-dimensional Mean-Field Games by Particle-based Flow Matching
- Flow Matching for Tabular Data Synthesis
- BIFE: Better Interaction, Fewer Errors for Minute-Long Video Generation
- Risk-Entropic Flow Matching
- Overcoming the Curvature Bottleneck in MeanFlow
- Generative Anchored Fields: Controlled Data Generation via Emergent Velocity Fields and Transport Algebra
- Test-time scaling of diffusions with flow maps
- Flow Density Control: Generative Optimization Beyond Entropy-Regularized Fine-Tuning
- CaFlow: Enhancing Long-Term Action Quality Assessment with Causal Counterfactual Flow
- Which Layer Causes Distribution Deviation? Entropy-Guided Adaptive Pruning for Diffusion and Flow Models
- MFM-point: Multi-scale Flow Matching for Point Cloud Generation
- Deep Parameter Interpolation for Scalar Conditioning
- Dynamic Test-Time Compute Scaling in Control Policy: Difficulty-Aware Stochastic Interpolant Policy
- TReFT: Taming Rectified Flow Models For One-Step Image Translation
- Flow Map Distillation Without Data
- Understanding, Accelerating, and Improving MeanFlow Training
- VeCoR -- Velocity Contrastive Regularization for Flow Matching
- Breaking the Likelihood-Quality Trade-off in Diffusion Models by Merging Pretrained Experts
- TRANSPORTER: Transferring Visual Semantics from VLM Manifolds
- TriDiff-4D: Fast 4D Generation through Diffusion-based Triplane Re-posing
- SURFing to the Fundamental Limit of Jet Tagging
- BokehFlow: Depth-Free Controllable Bokeh Rendering via Flow Matching
- Learning Straight Flows: Variational Flow Matching for Efficient Generation
- Learning by Neighbor-Aware Semantics, Deciding by Open-form Flows: Towards Robust Zero-Shot Skeleton Action Recognition
- Rectified Noise: A Generative Model Using Positive-incentive Noise
- Controllable Flow Matching for Online Reinforcement Learning
- On Flow Matching KL Divergence
- SAD-Flower: Flow Matching for Safe, Admissible, and Dynamically Consistent Planning
- Knothe-Rosenblatt maps via soft-constrained optimal transport
- Learning Paths for Dynamic Measure Transport: A Control Perspective
- Modeling Microenvironment Trajectories on Spatial Transcriptomics with NicheFlow
- Simplex-to-Euclidean Bijections for Categorical Flow Matching
- Curly Flow Matching for Learning Non-gradient Field Dynamics
- Efficient Generative AI Boosts Probabilistic Forecasting of Sudden Stratospheric Warmings
- SplitFlow: Flow Decomposition for Inversion-Free Text-to-Image Editing
- Flow Map Learning via Nongradient Vector Flow
- Autoregressive Boltzmann Generators
- Surflo: Consistent 3D Surface Flow Model with Global State
- Protein Language Model Embeddings Improve Generalization of Implicit Transfer Operators
- Meta Flow Maps enable scalable reward alignment
- LLM Meets Diffusion: A Hybrid Framework for Crystal Material Generation
- FlowCritic: Bridging Value Estimation with Flow Matching in Reinforcement Learning
- FlowOpt: Fast Optimization Through Whole Flow Processes for Training-Free Editing
- Generalised Flow Maps for Few-Step Generative Modelling on Riemannian Manifolds
- Epipolar Geometry Improves Video Generation Models
- Exponential Convergence Guarantees for Iterative Markovian Fitting
- AlphaFlow: Understanding and Improving MeanFlow Models
- FLOWR.root: A flow matching based foundation model for joint multi-purpose structure-aware 3D ligand generation and affinity prediction
- GuideFlow3D: Optimization-Guided Rectified Flow For Appearance Transfer
- LeapFactual: Reliable Visual Counterfactual Explanation Using Conditional Flow Matching
- Exploring Cross-Modal Flows for Few-Shot Learning
- pi-Flow: Policy-Based Few-Step Generation via Imitation Distillation
- Graph Generation with Spectral Geodesic Flow Matching
- VIST3A: Text-to-3D by Stitching a Multi-view Reconstruction Network to a Video Generator
- PAINT: Parallel-in-time Neural Twins for Dynamical System Reconstruction
- Equilibrium Matching: Generative Modeling with Implicit Energy-Based Models
- An Eulerian Perspective on Straight-Line Sampling
- Fine-Tuning Flow Matching via Maximum Likelihood Estimation of Reconstructions
- Asymmetric Flow Models
- Conditional Flow Matching for Bayesian Posterior Inference
- Mirror Flow Matching with Heavy-Tailed Priors for Generative Modeling on Convex Domains
- Counterfactual Identifiability via Dynamic Optimal Transport
- IntMeanFlow: Few-step Speech Generation with Integral Velocity Distillation
- Deep Neural Networks Inspired by Differential Equations
- Value Flows
- RePainter: Empowering E-commerce Object Removal via Spatial-matting Reinforcement Learning
- Pixel-Perfect Depth with Semantics-Prompted Diffusion Transformers
- Three Forms of Stochastic Injection for Improved Distribution-to-Distribution Generative Modeling
- Fine-Tuning Masked Diffusion for Provable Self-Correction
- Fundamental Limits of Crystalline Equivariant Graph Neural Networks: A Circuit Complexity Perspective
- StaMo: Unsupervised Learning of Generalizable Robot Motion from Compact State Representation
- SSDD: Single-Step Diffusion Decoder for Efficient Image Tokenization
- Flow-Matching Based Refiner for Molecular Conformer Generation
- ChronoEdit: Towards Temporal Reasoning for Image Editing and World Simulation
- ContextFlow: Context-Aware Flow Matching For Trajectory Inference From Spatial Omics Data
- Deep Generative Continual Learning using Functional LoRA: FunLoRA
- Quantifying the noise sensitivity of the Wasserstein metric for images
- Continuously Augmented Discrete Diffusion model for Categorical Generative Modeling
- Flow Autoencoders are Effective Protein Tokenizers
- Stitch: Training-Free Position Control in Multimodal Diffusion Transformers
- FLOWER: A Flow-Matching Solver for Inverse Problems
- Editable Noise Map Inversion: Encoding Target-image into Noise For High-Fidelity Image Manipulation
- Reframing Generative Models for Physical Systems using Stochastic Interpolants
- MarS-FM: Generative Modeling of Molecular Dynamics via Markov State Models
- Diffusion Bridge or Flow Matching? A Unifying Framework and Comparative Analysis
- Space Group Conditional Flow Matching
- VMDiff: Visual Mixing Diffusion for Limitless Cross-Object Synthesis
- Stochastic Interpolants via Conditional Dependent Coupling
- Generation Properties of Stochastic Interpolation under Finite Training Set
- Taming Flow-based I2V Models for Creative Video Editing
- Discrete Guidance Matching: Exact Guidance for Discrete Flow Matching
- Error Analysis of Discrete Flow with Generator Matching
- DriftLite: Lightweight Drift Control for Inference-Time Scaling of Diffusion Models
- Matérn Noise for Triangulation-Agnostic Flow Matching on Meshes
- ELF: Embedded Language Flows
- Self-Supervised Flow Matching for Scalable Multi-Modal Synthesis
- Multi-domain Distribution Learning for De Novo Drug Design
- CAR-Flow: Condition-Aware Reparameterization Aligns Source and Target for Better Flow Matching
- Masked Diffusion Models as Energy Minimization
- Flow-Based Fragment Identification via Binding Site-Specific Latent Representations
- LADB: Latent Aligned Diffusion Bridges for Semi-Supervised Domain Translation
- floq: Training Critics via Flow-Matching for Scaling Compute in Value-Based RL
- FLOWER: Democratizing Generalist Robot Policies with Efficient Vision-Language-Action Flow Policies
- Equivariant Flow Matching for Symmetry-Breaking Bifurcation Problems
- Scale-Adaptive Generative Flows for Multiscale Scientific Data
- Distribution estimation via Flow Matching with Lipschitz guarantees
- Multimodal Generative Flows for LHC Jets
- Lipschitz-Guided Design of Interpolation Schedules in Generative Models
- Challenges in Non-Polymeric Crystal Structure Prediction: Why a Geometric, Permutation-Invariant Loss is Needed
- Any-Order Flexible Length Masked Diffusion
- Quantum latent distributions in deep generative models
- MotionFlux: Efficient Text-Guided Motion Generation through Rectified Flow Matching and Preference Alignment
- Energy-Based Flow Matching for Generating 3D Molecular Structure
- Provable Mixed-Noise Learning with Flow-Matching
- Tinker: Diffusion's Gift to 3D--Multi-View Consistent Editing From Sparse Inputs without Per-Scene Optimization
- Generative Super-Resolution of Turbulent Flows via Stochastic Interpolants
- FlowMol3: Flow Matching for 3D De Novo Small-Molecule Generation
- Constraint-Aware Flow Matching via Randomized Exploration
- Flow Matching for Efficient and Scalable Data Assimilation
- Noise Hypernetworks: Amortizing Test-Time Compute in Diffusion Models
- When and how can inexact generative models still sample from the data manifold?
- Unsupervised Real-World Super-Resolution via Rectified Flow Degradation Modelling
- Finite-Time Convergence Analysis of ODE-based Generative Models for Stochastic Interpolants
- Towards High-Order Mean Flow Generative Models: Feasibility, Expressivity, and Provably Efficient Criteria
- OM2P: Offline Multi-Agent Mean-Flow Policy
- Multitask Learning with Stochastic Interpolants
- Bridging Diffusion Models and 3D Representations: A 3D Consistent Super-Resolution Framework
- SCFlow: Implicitly Learning Style and Content Disentanglement with Flow Models
- Tackling Ill-posedness of Reversible Image Conversion with Well-posed Invertible Network
- VFP: Variational Flow-Matching Policy for Multi-Modal Robot Manipulation
- Flow Matching for Probabilistic Learning of Dynamical Systems from Missing or Noisy Data
- One-Step Flow Policy Mirror Descent
- FMIP: Joint Continuous-Integer Flow For Mixed-Integer Linear Programming
Discussions
Related