pith. machine review for the scientific record. sign in

arxiv: 2303.08797 · v4 · submitted 2023-03-15 · 💻 cs.LG · cond-mat.dis-nn· math.PR

Recognition: 3 theorem links

· Lean Theorem

Stochastic Interpolants: A Unifying Framework for Flows and Diffusions

Eric Vanden-Eijnden, Michael S. Albergo, Nicholas M. Boffi

Pith reviewed 2026-05-11 20:38 UTC · model grok-4.3

classification 💻 cs.LG cond-mat.dis-nnmath.PR
keywords stochastic interpolantsgenerative modelsflow modelsdiffusion modelsFokker-Planck equationsSchrödinger bridgeprobability flowscore matching
0
0 comments X

The pith

Stochastic interpolants unify flow-based and diffusion-based generative models by exactly bridging any two probability densities in finite time.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a class of generative models that use stochastic interpolants to connect any two prescribed probability densities. These interpolants are built from samples of the two densities plus a latent variable that flexibly shapes the path. The time-dependent density obeys a transport equation together with a family of forward and backward Fokker-Planck equations whose diffusion coefficient is adjustable. This viewpoint produces both deterministic probability-flow ODEs and stochastic differential equations for generation, where the required drift or velocity fields arise as the unique minimizers of simple quadratic objectives. The construction recovers score-based diffusion models, rectifying flows, and the Schrödinger bridge as special cases while supplying estimators for likelihood and cross-entropy.

Core claim

A broad class of continuous-time stochastic processes called stochastic interpolants bridges any two probability density functions exactly in finite time. The interpolants combine data from the two target densities with an additional latent variable. Their time-dependent density satisfies both a transport equation and a family of Fokker-Planck equations with tunable diffusion coefficient. The associated drift coefficients are characterized as the unique minimizers of quadratic objective functions, one of which is new for the score, yielding deterministic and stochastic generative models.

What carries the argument

The stochastic interpolant, a continuous-time process formed from two target densities and a latent variable, whose drift fields are the unique minimizers of quadratic objectives obtained from the transport and Fokker-Planck equations.

If this is right

  • Minimization of the quadratic objectives controls the likelihood for generative models built on stochastic dynamics.
  • Likelihood control for deterministic dynamics is more stringent than for stochastic dynamics.
  • Explicit optimization over the interpolant recovers the Schrödinger bridge between the two target densities.
  • Estimators exist for the likelihood and cross-entropy of interpolant-based generative models.
  • The framework recovers score-based diffusion models, stochastic localization, probabilistic denoising, and rectifying flows as special cases.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The tunable diffusion coefficient could be used to adapt the model to data sets with different degrees of multimodality or noise.
  • Hybrid training that switches between the deterministic and stochastic formulations might improve sample quality or training speed in high dimensions.
  • Different choices of the latent variable could produce new families of exact bridges beyond the ones examined in the paper.

Load-bearing premise

The time-dependent density of the interpolant satisfies both the transport equation and the family of Fokker-Planck equations with the chosen diffusion coefficient, so that the resulting drift fields are exactly the unique minimizers of the stated quadratic objectives.

What would settle it

A numerical experiment in which samples generated by the learned probability-flow ODE or SDE starting from the source density fail to match the target density, for example when both densities are simple Gaussians.

read the original abstract

A class of generative models that unifies flow-based and diffusion-based methods is introduced. These models extend the framework proposed in Albergo and Vanden-Eijnden (2023), enabling the use of a broad class of continuous-time stochastic processes called stochastic interpolants to bridge any two probability density functions exactly in finite time. These interpolants are built by combining data from the two prescribed densities with an additional latent variable that shapes the bridge in a flexible way. The time-dependent density function of the interpolant is shown to satisfy a transport equation as well as a family of forward and backward Fokker-Planck equations with tunable diffusion coefficient. Upon consideration of the time evolution of an individual sample, this viewpoint leads to both deterministic and stochastic generative models based on probability flow equations or stochastic differential equations with an adjustable level of noise. The drift coefficients entering these models are time-dependent velocity fields characterized as the unique minimizers of simple quadratic objective functions, one of which is a new objective for the score. We show that minimization of these quadratic objectives leads to control of the likelihood for generative models built upon stochastic dynamics, while likelihood control for deterministic dynamics is more stringent. We also construct estimators for the likelihood and the cross entropy of interpolant-based generative models, and we discuss connections with other methods such as score-based diffusion models, stochastic localization, probabilistic denoising, and rectifying flows. In addition, we demonstrate that stochastic interpolants recover the Schr\"odinger bridge between the two target densities when explicitly optimizing over the interpolant. Finally, algorithmic aspects are discussed and the approach is illustrated on numerical examples.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 3 minor

Summary. The paper introduces stochastic interpolants as a unifying framework for flow-based and diffusion-based generative models. These are constructed by combining samples from two target densities with a latent variable to form continuous-time stochastic processes that exactly bridge the densities in finite time. The marginal density of the interpolant is shown to satisfy a transport equation and a tunable family of forward/backward Fokker-Planck equations. This yields both deterministic probability-flow ODEs and stochastic SDEs for generation, with time-dependent drift fields characterized as the unique minimizers of quadratic objectives (including a novel score objective). The work establishes likelihood control via these objectives, constructs estimators for likelihood and cross-entropy, recovers the Schrödinger bridge as a special case by optimizing the interpolant, discusses connections to score-based diffusion models and related methods, and provides numerical illustrations.

Significance. If the central derivations hold, the framework is significant for providing a flexible, self-contained way to interpolate between arbitrary densities with exact finite-time matching and controllable noise levels. The quadratic objectives and their variational characterization as L2 projections offer a clean training approach that generalizes existing flow and diffusion methods, while the likelihood analysis and Schrödinger bridge recovery add theoretical value. The approach could facilitate new algorithms by allowing tunable stochasticity without sacrificing endpoint accuracy.

major comments (2)
  1. [quadratic objectives section] § on quadratic objectives and unique minimizers: The claim that the drifts are the unique minimizers of the stated quadratic objectives (including the new score objective) is load-bearing for the generative model construction. The variational argument via L2 projection onto the tangent space of the marginal flow is sketched, but the precise function space (e.g., weighted L2) and regularity conditions on the interpolant and latent variable that guarantee strict convexity or uniqueness are not fully specified; this needs explicit statement to support the uniqueness for general choices.
  2. [likelihood control section] § on likelihood control: The distinction that minimization controls likelihood for stochastic dynamics (via Girsanov/Fokker-Planck) but is more stringent for deterministic dynamics is central to the comparison with flows. A concrete quantitative bound or simple counterexample illustrating when deterministic models lose likelihood control would strengthen the claim, as the current argument relies on standard change-of-measure formulas without detailing the gap.
minor comments (3)
  1. [introduction] The notation for the latent variable and its distribution could be introduced more explicitly in the opening paragraphs to aid readability before the formal definitions.
  2. [numerical examples] In the numerical examples, additional details on the specific diffusion coefficient schedules used and their sensitivity would help readers reproduce and assess the tunable noise aspect.
  3. [connections section] A few citations to recent work on rectifying flows and stochastic localization appear abbreviated; expanding the discussion paragraph would better situate the contributions.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading, positive summary, and recommendation for minor revision. We address each major comment below.

read point-by-point responses

  1. Referee: [quadratic objectives section] § on quadratic objectives and unique minimizers: The claim that the drifts are the unique minimizers of the stated quadratic objectives (including the new score objective) is load-bearing for the generative model construction. The variational argument via L2 projection onto the tangent space of the marginal flow is sketched, but the precise function space (e.g., weighted L2) and regularity conditions on the interpolant and latent variable that guarantee strict convexity or uniqueness are not fully specified; this needs explicit statement to support the uniqueness for general choices.

    Authors: We appreciate the referee's observation. The quadratic objectives are defined over the weighted L^2 space of vector fields with respect to the marginal density rho_t of the interpolant (i.e., integrable v such that integral |v|^2 rho_t dx < infinity). The objective is a strictly convex quadratic functional on this Hilbert space, so the minimizer is unique whenever the objective is finite. This holds under the mild regularity conditions already implicit in the construction (finite second moments of the interpolant and latent variable, with the interpolant being Lipschitz in time). We will add an explicit statement of the function space and these conditions to the quadratic objectives section. revision: yes

  2. Referee: [likelihood control section] § on likelihood control: The distinction that minimization controls likelihood for stochastic dynamics (via Girsanov/Fokker-Planck) but is more stringent for deterministic dynamics is central to the comparison with flows. A concrete quantitative bound or simple counterexample illustrating when deterministic models lose likelihood control would strengthen the claim, as the current argument relies on standard change-of-measure formulas without detailing the gap.

    Authors: We agree that a concrete illustration would strengthen the presentation. In the deterministic case the change-of-variables formula for the likelihood includes an additional integral of the divergence of the velocity field, which is not directly controlled by the quadratic objective (unlike the Girsanov correction for the stochastic case). A simple 1D counterexample is a constant-velocity flow versus a linearly compressing flow that achieve identical quadratic losses yet produce different log-likelihoods due to the Jacobian determinant. We will insert a short remark with this example and a brief quantitative comparison in the likelihood control section. revision: yes

Circularity Check

0 steps flagged

Minor self-citation of prior framework; central derivations self-contained from SDE and variational principles

full rationale

The paper extends Albergo & Vanden-Eijnden (2023) but derives the transport/Fokker-Planck equations, quadratic objectives for drifts, and likelihood controls directly from the stochastic interpolant definition, marginal flow, and standard Girsanov/Fokker-Planck change-of-measure formulas. No step reduces a claimed prediction or uniqueness result to a fitted parameter or self-citation chain; the variational characterization of velocity fields as L2 projections is independent of the target densities. The framework is therefore self-contained against external stochastic-process benchmarks, with only a non-load-bearing reference to the authors' prior interpolant construction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The framework rests on standard stochastic calculus (Fokker-Planck, SDEs) plus the new postulate that interpolants built from data plus latent variable exactly bridge the marginals and yield unique quadratic minimizers for the drifts.

free parameters (1)
  • diffusion coefficient schedule
    Tunable parameter appearing in the family of Fokker-Planck equations; controls noise level in the generative SDE.
axioms (1)
  • domain assumption The interpolant density satisfies a transport equation and a family of forward/backward Fokker-Planck equations.
    Invoked to derive both deterministic and stochastic generative dynamics.
invented entities (1)
  • stochastic interpolant no independent evidence
    purpose: Continuous-time process that exactly bridges two prescribed densities using data plus latent variable.
    Core new object introduced to unify flows and diffusions.

pith-pipeline@v0.9.0 · 5601 in / 1374 out tokens · 61762 ms · 2026-05-11T20:38:33.282090+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

Forward citations

Cited by 43 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Divergence is Uncertainty: A Closed-Form Posterior Covariance for Flow Matching

    cs.LG 2026-05 unverdicted novelty 8.0

    In flow matching, the uncertainty of the clean data given the current state is exactly the divergence of the velocity field (up to a known scalar).

  2. Flow-GRPO: Training Flow Matching Models via Online RL

    cs.CV 2025-05 unverdicted novelty 8.0

    Flow-GRPO is the first online RL method for flow matching models, raising GenEval accuracy from 63% to 95% and text-rendering accuracy from 59% to 92% with little reward hacking.

  3. Aligning Flow Map Policies with Optimal Q-Guidance

    cs.LG 2026-05 unverdicted novelty 7.0

    Flow map policies enable fast one-step inference for flow-based RL policies, and FMQ provides an optimal closed-form Q-guided target for offline-to-online adaptation under trust-region constraints, achieving SOTA performance.

  4. Generative Transfer for Entropic Optimal Transport with Unknown Costs

    math.OC 2026-05 unverdicted novelty 7.0

    A generative transfer framework using iterative path-wise tilting integrated with conditional flow matching recovers target entropic optimal transport couplings from reference samples, achieving O(δ) convergence in Wa...

  5. Is Monotonic Sampling Necessary in Diffusion Models?

    cs.LG 2026-05 unverdicted novelty 7.0

    Non-monotonic sampling schedules never improve upon monotonic baselines in diffusion models, with performance gaps ranging from substantial to negligible depending on the denoiser.

  6. TMPO: Trajectory Matching Policy Optimization for Diverse and Efficient Diffusion Alignment

    cs.LG 2026-05 unverdicted novelty 7.0

    TMPO replaces scalar reward maximization with trajectory-level matching to a Boltzmann distribution via Softmax-TB, improving generative diversity by 9.1% while keeping competitive reward performance.

  7. FLUX: Geometry-Aware Longitudinal Flow Matching with Mixture of Experts

    cs.LG 2026-05 unverdicted novelty 7.0

    FLUX reconstructs longitudinal transport and recovers interpretable regime structure from unpaired biological snapshots by combining geometry-aware flow matching with mixture-of-experts velocity decomposition.

  8. Geometry-Aware Discretization Error of Diffusion Models

    cs.LG 2026-05 unverdicted novelty 7.0

    First-order asymptotic expansions of weak and Fréchet discretization errors in diffusion sampling are derived, explicit under Gaussian data through covariance geometry and robust to other data geometries.

  9. SDFlow: Similarity-Driven Flow Matching for Time Series Generation

    cs.AI 2026-05 unverdicted novelty 7.0

    SDFlow uses similarity-driven flow matching with low-rank manifold decomposition and a categorical posterior to generate high-fidelity long time series in VQ space without step-wise error accumulation.

  10. Bayesian Rain Field Reconstruction using Commercial Microwave Links and Diffusion Model Priors

    cs.LG 2026-05 unverdicted novelty 7.0

    Diffusion model priors enable training-free Bayesian sampling for more accurate rain field reconstruction from path-integrated commercial microwave link measurements than Gaussian process baselines.

  11. Generative Modeling with Orbit-Space Particle Flow Matching

    cs.GR 2026-05 unverdicted novelty 7.0

    OGPP is a particle flow-matching method using orbit-space canonicalization and geometric paths that achieves lower error and fewer steps than prior approaches on 3D benchmarks.

  12. ABC: Any-Subset Autoregression via Non-Markovian Diffusion Bridges in Continuous Time and Space

    cs.LG 2026-04 unverdicted novelty 7.0

    ABC enables any-subset autoregressive generation of continuous stochastic processes via non-Markovian diffusion bridges that track physical time and allow path-dependent conditioning.

  13. How to Guide Your Flow: Few-Step Alignment via Flow Map Reward Guidance

    cs.LG 2026-04 unverdicted novelty 7.0

    FMRG is a training-free, single-trajectory guidance method for flow models derived from optimal control that achieves strong reward alignment with only 3 NFEs.

  14. Annealed Langevin Monte Carlo for Flow ODE Sampling

    stat.CO 2026-04 unverdicted novelty 7.0

    ALMC-ODE uses annealed Langevin Monte Carlo with Jarzynski reweighting to produce a low-variance velocity estimator for flow ODE sampling, with an O(1/n) MSE bound and superior performance on multimodal benchmarks.

  15. ScoRe-Flow: Complete Distributional Control via Score-Based Reinforcement Learning for Flow Matching

    cs.RO 2026-04 unverdicted novelty 7.0

    ScoRe-Flow achieves decoupled mean-variance control in stochastic flow matching by deriving a closed-form score for drift modulation plus learned variance, yielding faster RL convergence and higher success rates on lo...

  16. MixGRPO: Unlocking Flow-based GRPO Efficiency with Mixed ODE-SDE

    cs.AI 2025-07 unverdicted novelty 7.0

    MixGRPO speeds up GRPO for flow-based image generators by restricting SDE sampling and optimization to a sliding window while using ODE elsewhere, cutting training time by up to 71% with better alignment performance.

  17. The two clocks and the innovation window: When and how generative models learn rules

    cs.LG 2026-05 unverdicted novelty 6.0

    Generative models learn rules before memorizing data, creating an innovation window whose width depends on dataset size and rule complexity, observed in both diffusion and autoregressive architectures.

  18. Conservative Flows: A New Paradigm of Generative Models

    cs.LG 2026-05 unverdicted novelty 6.0

    Conservative flows generate by running probability-preserving stochastic dynamics initialized at data points rather than noise, using corrected Langevin or predictor-corrector mechanisms on top of any pretrained flow ...

  19. Flow Matching with Arbitrary Auxiliary Paths

    cs.LG 2026-05 unverdicted novelty 6.0

    AuxPath-FM extends flow matching to arbitrary auxiliary distributions while preserving the continuity equation and marginal training objective.

  20. Physical Fidelity Reconstruction via Improved Consistency-Distilled Flow Matching for Dynamical Systems

    cs.LG 2026-05 unverdicted novelty 6.0

    Distilled one-step consistency model from optimal-transport flow-matching teacher reconstructs high-fidelity dynamical system flows from low-fidelity data with 12x speedup, half the parameters, and 23.1% better SSIM t...

  21. SDFlow: Similarity-Driven Flow Matching for Time Series Generation

    cs.AI 2026-05 unverdicted novelty 6.0

    SDFlow learns a global transport map via similarity-driven flow matching in VQ latent space, using low-rank manifold decomposition and a categorical posterior to handle discreteness, yielding SOTA long-horizon perform...

  22. Energy Generative Modeling: A Lyapunov-based Energy Matching Perspective

    cs.LG 2026-05 unverdicted novelty 6.0

    Training and sampling in static scalar energy generative models are two instances of the same Lyapunov-driven density transport dynamics on Wasserstein space, differing only by initial condition, which yields a finite...

  23. OGPO: Sample Efficient Full-Finetuning of Generative Control Policies

    cs.LG 2026-05 unverdicted novelty 6.0

    OGPO is a sample-efficient off-policy method for full finetuning of generative control policies that reaches SOTA on robotic manipulation tasks and can recover from poor behavior-cloning initializations without expert data.

  24. CARD: Coarse-to-fine Autoregressive Modeling with Radix-based Decomposition for Transferable Free Energy Estimation

    cs.LG 2026-05 unverdicted novelty 6.0

    CARD uses radix decomposition to enable autoregressive modeling of molecular coordinates as a zero-free-energy reference distribution, delivering classical accuracy for absolute free energy on unseen systems at ~40x speedup.

  25. Quotient-Space Diffusion Models

    cs.LG 2026-04 unverdicted novelty 6.0

    Quotient-space diffusion models handle symmetries by diffusing on the space of equivalent configurations under group actions like SE(3), reducing learning complexity and guaranteeing correct sampling for molecular generation.

  26. Uncertainty-Aware Spatiotemporal Super-Resolution Data Assimilation with Diffusion Models

    physics.flu-dyn 2026-04 unverdicted novelty 6.0

    DiffSRDA uses denoising diffusion models to perform uncertainty-aware spatiotemporal super-resolution data assimilation, achieving EnKF-like quality from low-resolution forecasts on an ocean jet testbed.

  27. Allo{SR}$^2$: Rectifying One-Step Super-Resolution to Stay Real via Allomorphic Generative Flows

    cs.CV 2026-04 unverdicted novelty 6.0

    Allo{SR}^2 rectifies one-step super-resolution trajectories with allomorphic generative flows via SNR initialization, velocity supervision, and self-adversarial matching to deliver state-of-the-art fidelity and realism.

  28. Denoising, Fast and Slow: Difficulty-Aware Adaptive Sampling for Image Generation

    cs.CV 2026-04 unverdicted novelty 6.0

    Patch Forcing enables diffusion models to denoise image patches at varying rates based on predicted difficulty, advancing easier regions first to improve context and achieve better generation quality on ImageNet while...

  29. Perron-Frobenius Contractive Operator Matching for Data-Driven Reachable Fault Identification and Recovery

    eess.SY 2026-04 unverdicted novelty 6.0

    A framework learns fault-indexed Perron-Frobenius operators from trajectory data to provide certifiable 2-Wasserstein bounds for detecting actuator faults and enabling recovery via density propagation in nonlinear con...

  30. PRiMeFlow: Capturing Complex Expression Heterogeneity in Perturbation Response Modelling

    cs.LG 2026-04 unverdicted novelty 6.0

    PRiMeFlow is a flow-matching model that approximates the full empirical distribution of single-cell gene expression after perturbations.

  31. MENO: MeanFlow-Enhanced Neural Operators for Dynamical Systems

    cs.LG 2026-04 unverdicted novelty 6.0

    MENO enhances neural operators with MeanFlow to restore multi-scale accuracy in dynamical system predictions while keeping inference costs low, achieving up to 2x better power spectrum accuracy and 12x faster inferenc...

  32. Improving Controllable Generation: Faster Training and Better Performance via $x_0$-Supervision

    cs.CV 2026-04 unverdicted novelty 6.0

    x0-supervision or equivalent loss re-weighting accelerates convergence in controllable diffusion models while improving visual quality and conditioning accuracy.

  33. Monte Carlo Event Generation with Continuous Normalizing Flows

    hep-ph 2026-04 conditional novelty 6.0

    Continuous normalizing flows improve unweighting efficiency in Monte Carlo event generation for high-jet-multiplicity collider processes by factors up to 184, with wall-time gains of about ten when combined with coupl...

  34. World Action Models are Zero-shot Policies

    cs.RO 2026-02 unverdicted novelty 6.0

    DreamZero uses a 14B video diffusion model as a World Action Model to achieve over 2x better zero-shot generalization on real robots than state-of-the-art VLAs, real-time 7Hz closed-loop control, and cross-embodiment ...

  35. DanceGRPO: Unleashing GRPO on Visual Generation

    cs.CV 2025-05 unverdicted novelty 6.0

    DanceGRPO applies GRPO to visual generation tasks to achieve stable policy optimization across diffusion models, rectified flows, multiple tasks, and diverse reward models, outperforming prior RL methods.

  36. Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think

    cs.CV 2024-10 unverdicted novelty 6.0

    Aligning noisy hidden states in diffusion transformers to clean features from pretrained visual encoders speeds up training over 17x and reaches FID 1.42.

  37. Sharpen Your Flow: Sharpness-Aware Sampling for Flow Matching

    cs.LG 2026-05 unverdicted novelty 5.0

    SharpEuler estimates a sharpness profile via finite differences on calibration trajectories, smooths it, and applies a quantile transform to generate adaptive timestep grids that improve Euler sampling quality in flow...

  38. Neural Posterior Estimation of Terrain Parameters from Radar Sounder Data

    eess.SP 2026-05 unverdicted novelty 5.0

    Neural posterior estimation trained on simulated radar data enables probabilistic inference of terrain parameters from real Mars radar sounder profiles while conditioning on reference surface assumptions.

  39. PRiMeFlow: Capturing Complex Expression Heterogeneity in Perturbation Response Modelling

    cs.LG 2026-04 unverdicted novelty 5.0

    PRiMeFlow applies flow matching in gene expression space with a U-Net velocity field and pretraining-finetuning to model perturbation-induced heterogeneity, showing strong benchmark performance on PerturBench and the ...

  40. Energy-oriented Diffusion Bridge for Image Restoration with Foundational Diffusion Models

    cs.CV 2026-04 unverdicted novelty 5.0

    E-Bridge approximates low-cost geodesic trajectories in diffusion bridges for image restoration by using shorter time horizons, entropy-regularized starts mixing degraded images with noise, and consistency-model singl...

  41. A Unified Measure-Theoretic View of Diffusion, Score-Based, and Flow Matching Generative Models

    cs.LG 2026-05 unverdicted novelty 4.0

    Diffusion, score-based, and flow matching models are unified as instances of learning time-dependent vector fields inducing marginal distributions governed by continuity and Fokker-Planck equations.

  42. Generative models for decision-making under distributional shift

    cs.LG 2026-04 unverdicted novelty 3.0

    Generative models via pushforward maps, Fokker-Planck equations, and Wasserstein geometry enable learning nominal uncertainty, stressed distributions for robustness, and conditional posteriors under distributional shift.

  43. Machine Learning Techniques for Astrophysics and Cosmology: Simulation-Based Inference

    astro-ph.CO 2026-05 unverdicted novelty 2.0

    Simulation-based inference uses neural networks trained on simulations to enable parameter inference in cosmology and astrophysics where traditional likelihood calculations are intractable.

Reference graph

Works this paper leans on

14 extracted references · 14 canonical work pages · cited by 41 Pith papers · 1 internal anchor

  1. [1]

    Diffusions hypercontractives

    Dominique Bakry and Michel ´Emery. Diffusions hypercontractives. In Jacques Az´ ema and Marc Yor, editors,S´ eminaire de Probabilit´ es XIX 1983/84, pages 177–206, Berlin, Heidelberg,

    work page 1983

  2. [2]

    ISBN 978-3-540-39397-9

    Springer Berlin Heidelberg. ISBN 978-3-540-39397-9. Heli Ben-Hamu, Samuel Cohen, Joey Bose, Brandon Amos, Maximilian Nickel, Aditya Grover, Ricky T. Q. Chen, and Yaron Lipman. Matching normalizing flows and probability paths on manifolds. InInternational Conference on Machine Learning, ICML 2022, 17-23 July 2022, Baltimore, Maryland, USA, volume 162 ofPro...

    work page 2022

  3. [3]

    The Schr \”odinger Bridge between Gaussian Measures has a Closed Form.arXiv:2202.05722, February

    Charlotte Bunne, Ya-Ping Hsieh, Marco Cuturi, and Andreas Krause. The Schr \”odinger Bridge between Gaussian Measures has a Closed Form.arXiv:2202.05722, February

    work page arXiv

  4. [4]

    simple diffusion: End-to-end diffusion for high resolution images

    Emiel Hoogeboom, Jonathan Heek, and Tim Salimans. simple diffusion: End-to-end diffusion for high resolution images. InInternational Conference on Machine Learning, ICML 2023, 23-29 July 2023, Honolulu, Hawaii, USA,

    work page 2023

  5. [5]

    Neural au- toregressive flows

    Chin-Wei Huang, David Krueger, Alexandre Lacoste, and Aaron Courville. Neural au- toregressive flows. In Jennifer Dy and Andreas Krause, editors,Proceedings of the 35th International Conference on Machine Learning, volume 80 ofProceedings of Machine Learning Research, pages 2078–2087. PMLR, 10–15 Jul

    work page 2078

  6. [6]

    Solving linear inverse problems using the prior implicit in a denoiser

    Zahra Kadkhodaie and Eero P. Simoncelli. Solving linear inverse problems using the prior implicit in a denoiser.arXiv:2007.13640,

    work page arXiv 2007

  7. [7]

    Convergence for score-based generative modeling with polynomial complexity.arXiv:2206.06227,

    Holden Lee, Jianfeng Lu, and Yixin Tan. Convergence for score-based generative modeling with polynomial complexity.arXiv:2206.06227,

    work page arXiv

  8. [8]

    I2SB: Image-to-image Schrödinger bridge.arXiv preprint arXiv:2302.05872,

    Guan-Horng Liu, Arash Vahdat, De-An Huang, Evangelos A Theodorou, Weili Nie, and Anima Anandkumar. I2sb: Image-to-image schr¨ odinger bridge.arXiv preprint arXiv:2302.05872, 2023a. Qiang Liu. Rectified flow: A marginal preserving approach to optimal transport. arXiv:2209.14577,

    work page arXiv

  9. [9]

    Let us build bridges: Understanding and extending diffusion generative models

    Xingchao Liu, Lemeng Wu, Mao Ye, and Qiang Liu. Let us build bridges: Understanding and extending diffusion generative models.arXiv:2208.14699,

    work page arXiv

  10. [10]

    cs.LG", eprint =

    77 Albergo, Boffi, and V anden-Eijnden Andrea Montanari. Sampling, diffusions, and stochastic localization.arXiv:2305.10690,

    work page arXiv

  11. [11]

    78 Stochastic Interpolants Yang Song and Diederik P. Kingma. How to train your energy-based models. arXiv:2101.03288,

    work page arXiv

  12. [12]

    Improving and generalizing flow-based generative models with minibatch optimal transport

    Alexander Tong, Nikolay Malkin, Guillaume Huguet, Yanlei Zhang, Jarrid Rector-Brooks, Kilian Fatras, Guy Wolf, and Yoshua Bengio. Conditional flow matching: Simulation-free dynamic optimal transport.arXiv:2302.00482,

    work page internal anchor Pith review arXiv

  13. [13]

    Vempala and Andre Wibisono

    Santosh S. Vempala and Andre Wibisono. Rapid convergence of the unadjusted langevin algorithm: Log-sobolev suffices.arXiv:1903.08568,

    work page arXiv 1903

  14. [14]

    Zhang, W

    Linfeng Zhang, Weinan E, and Lei Wang. Monge-Amp` ere flow for generative modeling. arXiv.org:1809.10188,

    work page arXiv