Diffusion differentiable resampling

2025/12/11 by Andersson, Jennifer Rosina, Zhao, Zheng
#FOS: Computer and information sciences #FOS: Mathematics #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Statistics Theory (math.ST)

paper · doi:10.48550/arxiv.2512.10401

Abstract

This paper is concerned with differentiable resampling in the context of sequential Monte Carlo (e.g., particle filtering). We propose a new informative resampling method that is instantly pathwise differentiable, based on an ensemble score diffusion model. We prove that our diffusion resampling method provides a consistent estimate to the resampling distribution, and we show by experiments that it outperforms the state-of-the-art differentiable resampling methods when used for stochastic filtering and parameter estimation.

Citations

• Dimension-free error estimate for diffusion model and optimal scheduling
• PyDPF: A Python Package for Differentiable Particle Filtering
• Generative diffusion posterior sampling for informative likelihoods
• SDE Matching: Scalable and Simulation-Free Training of Latent Stochastic Differential Equations
• Temperature Optimization for Bayesian Deep Learning
• Taming Diffusion Models for Image Restoration: A Review
• Conditional sampling within generative diffusion models
• Score matching for bridges without learning time-reversals
• Conditioning diffusion models by explicit forward-backward bridging
• Conditioning non-linear and infinite-dimensional diffusion processes
• Amortized Reparametrization: Efficient and Scalable Variational Inference for Latent SDEs
• On Feynman--Kac training of partial Bayesian neural networks
• An Ensemble Score Filter for Tracking High-Dimensional Nonlinear Dynamical Systems
• Stochastic filtering with moment representation
• Stochastic Interpolants: A Unifying Framework for Flows and Diffusions
• An overview of differentiable particle filters for data-adaptive sequential Bayesian inference
• Symbolic Discovery of Optimization Algorithms
• Image Restoration with Mean-Reverting Stochastic Differential Equations
• Improved Rate of First Order Algorithms for Entropic Optimal Transport
• Do Bayesian Neural Networks Need To Be Fully Stochastic?
• DPM-Solver++: Fast Solver for Guided Sampling of Diffusion Probabilistic Models
• Automatic Differentiation of Programs with Discrete Randomness
• Sampling is as easy as learning the score: theory for diffusion models with minimal data assumptions
• Convergence of denoising diffusion models under the manifold hypothesis
• Fast Sampling of Diffusion Models with Exponential Integrator
• On resampling schemes for particle filters with weakly informative\n observations
• On Neural Differential Equations
• Optimal Transport Tools (OTT): A JAX Toolbox for all things Wasserstein
• Variational Marginal Particle Filters
• Differentiable Particle Filtering without Modifying the Forward Pass
• Diffusion Schr "odinger Bridge with Applications to Score-Based\n Generative Modeling
• Efficient and Modular Implicit Differentiation
• Efficient and Accurate Gradients for Neural SDEs
• Differentiable Particle Filtering via Entropy-Regularized Optimal Transport
• Score-Based Generative Modeling through Stochastic Differential Equations
• Denoising Diffusion Probabilistic Models
• Implicit Neural Representations with Periodic Activation Functions
• Towards Differentiable Resampling
• Modeling Continuous Stochastic Processes with Dynamic Normalizing Flows
• Scalable Gradients for Stochastic Differential Equations
• Hamiltonian Neural Networks
• Analysis of the BFGS Method with Errors
• Particle Filter Networks with Application to Visual Localization
• Variational Sequential Monte Carlo
• Stein Variational Gradient Descent as Gradient Flow
• Importance Sampling: Intrinsic Dimension and Computational Cost
• An Overview on the Estimation of Large Covariance and Precision Matrices
• Theoretical guarantees for approximate sampling from smooth and log-concave densities
• Feedback Particle Filter
• A non-parametric ensemble transform method for Bayesian inference
• Toward Practical N2 Monte Carlo: the Marginal Particle Filter

Related

• Resampling Gradients Vanish in Differentiable Sequential Monte Carlo Samplers
• Variance Reduction of Resampling for Sequential Monte Carlo
• Stratification and Optimal Resampling for Sequential Monte Carlo
• Reverse Diffusion Sequential Monte Carlo Samplers
• Sequential quasi-Monte Carlo: Introduction for Non-Experts, Dimension\n Reduction, Application to Partly Observed Diffusion Processes