New method controls bias in unadjusted Hamiltonian Monte Carlo and underdamped Langevin.
problem Bias in unadjusted Hamiltonian Monte Carlo and underdamped Langevin samplers.
method Delocalization of bias technique applied to these samplers.
result Control W 2 W_2 W 2 bias with O ( K ) O(\sqrt{K}) O ( K ) integration steps for high-dimensional distributions. New methods improve efficiency of sampling algorithms for complex systems.
problem Efficiently sampling from complex, high-dimensional probability distributions.
method Randomized Runge-Kutta-Nyström methods tailored for Hamiltonian flows.
result Quantitative 5 / 2 5/2 5/2 -order L 2 L^2 L 2 -accuracy in approximating Hamiltonian flows. In this paper, we provide new insights on the Unadjusted Langevin Algorithm. We show that this method can be formulated as a first order optimization algorithm of an objective functional defined on the Wasserstein space of order 2 2 2 . Using this interpretation and techniques borrowed from convex optimization, we give a …
The unadjusted Langevin algorithm converges faster for some variables in high dimensions.
problem Sampling probability distributions in high-dimensional settings.
method Analysis of the unadjusted Langevin algorithm for strongly log-concave distributions.
result The delocalization of bias effect allows for faster convergence for a small number of variables.
New sampling method for heavy-tailed distributions using Langevin Algorithm.
problem Sampling from heavy-tailed distributions efficiently.
method Transformed Unadjusted Langevin Algorithm on specific transformations.
result Polynomial-order oracle complexities for certain heavy-tailed densities.
ULA estimates covariance of log-concave distributions efficiently.
problem Estimating covariance matrices of log-concave distributions efficiently.
method Unadjusted Langevin algorithm (ULA) for sampling and covariance estimation.
result Sample complexity of single-chain ULA is smaller than that of parallel ULA by a logarithmic factor.
We study the Unadjusted Langevin Algorithm (ULA) for sampling from a probability distribution ν = e − f ν= e^{-f} ν = e − f on R n \mathbb{R}^n R n . We prove a convergence guarantee in Kullback-Leibler (KL) divergence assuming ν ν ν satisfies a log-Sobolev inequality and the Hessian of f f f is bounded. Notably, we do not assume convexity or boun…
MAFLA improves sampling from heavy-tailed distributions using MH-inspired corrections.
problem Sampling from heavy-tailed and multimodal distributions when neither target nor proposal densities can be evaluated.
method Metropolis-Adjusted Fractional Langevin Algorithm (MAFLA) with Score Balance Matching.
result MAFLA significantly improves finite-time sampling accuracy over unadjusted fractional Langevin dynamics.
The paper analyzes convergence rates of Langevin dynamics and Proximal Sampler using Φ Φ Φ -divergence.
problem Analyzing convergence rates of Langevin dynamics and Proximal Sampler.
method Extending mixing time analyses to Φ Φ Φ -divergence, using strong data processing inequalities. result Convergence of Φ Φ Φ -divergence to 0 exponentially fast along Unadjusted Langevin Algorithm and Proximal Sampler. New methods use transport maps to improve Langevin dynamics for sampling.
problem Sampling high-dimensional, non-Gaussian distributions efficiently.
method Apply transport maps to accelerate Langevin dynamics convergence.
result Discretized processes converge to target distribution with non-asymptotic bounds.
We study the Proximal Langevin Algorithm (PLA) for sampling from a probability distribution ν = e − f ν= e^{-f} ν = e − f on R n \mathbb{R}^n R n under isoperimetry. We prove a convergence guarantee for PLA in Kullback-Leibler (KL) divergence when ν ν ν satisfies log-Sobolev inequality (LSI) and f f f has bounded second and third derivatives. Thi…
Unified approach for sampling non-differentiable and heavy-tailed targets.
problem Sampling non-differentiable and heavy-tailed distributions using Langevin algorithms.
method Anchored Langevin dynamics, which modifies the Langevin diffusion with a smooth reference potential and multiplicative scaling.
result Non-asymptotic guarantees in the 2-Wasserstein distance to the target distribution.
New Langevin algorithm works well even for rough distributions.
problem Sampling from non-smooth distributions.
method Simple Langevin algorithm without smoothness assumptions.
result Algorithm performs well even with discontinuous gradients.
Proves convergence of PSGLA for sampling non-convex potentials.
problem Sampling from non-convex potentials with stability.
method Combines ULA and proximal optimization with stability analysis.
result First proof of convergence for PSGLA on non-convex potentials.
We study sampling as optimization in the space of measures. We focus on gradient flow-based optimization with the Langevin dynamics as a case study. We investigate the source of the bias of the unadjusted Langevin algorithm (ULA) in discrete time, and consider how to remove or reduce the bias. We point out the difficul…
Analyzes learning and applying preconditioners in MCMC for efficiency.
problem Improving efficiency of MCMC algorithms.
method Non-asymptotic analysis of schemes that learn preconditioners.
result Established non-asymptotic guarantees for preconditioned ULA.
The Riemannian Langevin Algorithm samples from manifolds efficiently.
problem Sampling from distributions on manifolds with log-Sobolev inequality.
method Riemannian Langevin Algorithm, log-Sobolev inequality, self-concordance extension, stochastic smoothness bounding.
result The Riemannian Langevin Algorithm converges rapidly to the target density.
New algorithm improves stability and efficiency of neural network training.
problem Stability and efficiency issues in adaptive optimization for neural networks.
method Polygonal approximations for SDEs with monotone coefficients, providing stability and addressing vanishing gradients.
result TheoPouLa algorithm shows superior performance over popular adaptive optimizers.
The paper provides privacy guarantees for MCMC algorithms using Langevin dynamics.
problem Ensuring differential privacy in MCMC algorithms.
method Novel methodology combining Girsanov's theorem and perturbation trick.
result Established (Rényi) DP guarantees for Langevin algorithms.
New algorithms improve sampling from constrained distributions.
problem Generating samples from distributions under constraints.
method Kinetic Langevin dynamics and splitting schemes.
result Improved complexity bounds over existing methods.
New schemes improve error estimates for sampling from non-log-concave distributions.
problem Improving sampling from non-log-concave distributions with super-linear drift growth.
method Developed tamed Euler and randomized Euler schemes with error estimates.
result Near-optimal error bounds for sampling and optimization problems.
New algorithms for sampling and optimization without tuning.
problem Efficient sampling and optimization over probability measures.
method Optimization on the space of probability measures, using gradient flows.
result Strong theoretical guarantees and similar performance to optimally tuned algorithms.
Stochastic EM with biased MCMC improves inference stability.
problem Intractable E-step in EM algorithm.
method Stochastic approximation with biased MCMC.
result ULA is more stable and sometimes faster than MALA.
We consider the problem of sampling from a strongly log-concave density in R d \mathbb{R}^d R d , and prove a non-asymptotic upper bound on the mixing time of the Metropolis-adjusted Langevin algorithm (MALA). The method draws samples by simulating a Markov chain obtained from the discretization of an appropriate Langevin dif…
New algorithm TUSLA improves learning of non-convex neural networks.
problem Optimizing non-convex loss functions in neural networks with superlinear gradient growth.
method Tamed Unadjusted Stochastic Langevin Algorithm (TUSLA) based on SGLD with taming technology.
result Finite-time guarantees for TUSLA to find approximate minimizers of empirical and population risks.
New method samples from non-log-concave distributions with weak dissipativity.
problem Sampling from distributions that are not log-concave and weakly dissipative.
method Taming scheme tailored to growth and decay properties of the target distribution.
result Explicit non-asymptotic guarantees for KL, TV, and Wasserstein distances.
NF-ULA combines Langevin Monte Carlo with normalizing flows for imaging inverse problems.
problem Solving inverse problems in imaging with uncertainty quantification.
method Langevin Monte Carlo with normalizing flow prior.
result NF-ULA outperforms competing methods for severely ill-posed inverse problems.
New method improves sampling from score-based models by correcting bias.
problem Bias in sampling from score-based diffusion models.
method Metropolis-Hastings or Barker's accept-reject steps to correct bias, using the score function.
result Improves sample quality on synthetic and image datasets, yielding consistent gains in FID.
New algorithm improves latent variable model estimation.
problem Estimating parameters in latent variable models.
method Jarzynski-adjusted Langevin algorithm (JALA) for SMC methods.
result JALA-EM provides maximum marginal likelihood estimate.
Markov chain (MC) algorithms are ubiquitous in machine learning and statistics and many other disciplines. Typically, these algorithms can be formulated as acceptance rejection methods. In this work we present a novel estimator applicable to these methods, dubbed Markov chain importance sampling (MCIS), which efficient…
A new sampler for complex discrete distributions efficiently updates all variables in parallel.
problem Sampling complex high-dimensional discrete distributions efficiently and accurately.
method Discrete Langevin proposal (DLP) for parallel coordinate updates with controlled stepsize.
result DLP efficiently explores high-dimensional and strongly correlated variables with asymptotic bias of zero for log-quadratic distributions.
LMC achieves sqrt(d) dependence in sampling error, improving previous bounds.
problem Analyzing sampling error in Langevin Monte Carlo.
method Refined mean-square analysis for discretizations of contractive SDEs.
result Establishes i l d e O ( d / ε ) ilde{O}(\sqrt{d}/ε) i l d e O ( d / ε ) mixing time bound for LMC. New method accelerates Bayesian imaging using Langevin sampling.
problem Bayesian inference in imaging inverse problems with convex geometry.
method Stochastic relaxed proximal-point iteration targeting posterior distribution.
result Accelerated convergence for κ κ κ -strongly log-concave targets. Statistical finite elements use Langevin dynamics to efficiently handle uncertainty quantification.
problem Uncertainty quantification in finite element models with observed data.
method Langevin dynamics, unadjusted Langevin algorithm (ULA), for sampling posterior distributions.
result ULA provides a scalable and efficient method for characterizing the posterior distribution of statFEM models.
Improved sampling from complex distributions with reduced bias.
problem Reducing bias in high-dimensional sampling algorithms.
method Hierarchical entropy analysis to weaken assumptions and expand scope.
result Bias reduction in low-dimensional marginals scales with lower dimension, not full dimension.
We study the problem of robustly estimating the posterior distribution for the setting where observed data can be contaminated with potentially adversarial outliers. We propose Rob-ULA, a robust variant of the Unadjusted Langevin Algorithm (ULA), and provide a finite-sample analysis of its sampling distribution. In par…
The paper studies how quickly samples from Langevin dynamics become independent.
problem Understanding the dependence between samples along Langevin dynamics and related algorithms.
method Measures dependence via Φ Φ Φ -mutual information and proves strong data processing inequalities. result The Φ Φ Φ -mutual information between samples decreases exponentially to zero. Generative model uses Schrödinger bridges for stable sampling.
problem Sampling from unknown distributions with limited training samples.
method Combines Schrödinger bridges and Langevin dynamics.
result Effective stability and generation of samples within convex hull.
This paper analyzes the bias of inexact MCMC methods in high dimensions.
problem Understanding the bias of inexact MCMC methods in high-dimensional spaces.
method Establishing bounds on Wasserstein distances between inexact MCMC methods and target distributions.
result The asymptotic bias of ULA and uHMC depends on key quantities related to the target distribution or the stationary probability measure of the scheme.
New method improves sampling efficiency in complex stochastic systems.
problem Sampling efficiency in nonconvex stochastic gradient cases.
method Reflection coupling for unadjusted generalized Hamiltonian Monte Carlo.
result Quantitative Gaussian concentration bounds and convergence rates established.
LMC algorithm converges to target in Chi-squared and Renyi divergence.
problem Sampling from target distribution using LMC with strong dissipativity and smoothness conditions.
method LMC algorithm with strong dissipativity and first-order smoothness, initialized with Gaussian.
result LMC reaches ε-neighborhood of target in Chi-squared and Renyi divergence in O(λ²dε⁻¹) steps.
Adaptive Langevin dynamics reduces bias in Bayesian inference with mini-batching.
problem Bias in posterior sampling due to mini-batching in Bayesian inference.
method Adaptive Langevin dynamics with dynamical friction to correct noise.
result Quantified bias in posterior distribution due to mini-batching.
Noise-free sampling method using Wasserstein proximal for faster convergence.
problem Sampling from distributions governed by potential functions.
method Deterministic score-based MCMC with regularized Wasserstein proximal.
result Improved mixing time bounds for Gaussian distributions compared to ULA and MALA.
We consider in this paper the problem of sampling a high-dimensional probability distribution π π π having a density with respect to the Lebesgue measure on R d \mathbb{R}^d R d , known up to a normalization constant x ↦ π ( x ) = e − U ( x ) / ∫ R d e − U ( y ) d y x \mapsto π(x)= \mathrm{e}^{-U(x)}/\int_{\mathbb{R}^d} \mathrm{e}^{-U(y)} \mathrm{d} y x ↦ π ( x ) = e − U ( x ) / ∫ R d e − U ( y ) d y . Such problem naturally…
New analysis for learning and applying preconditioners in MCMC improves efficiency.
problem Improving efficiency of MCMC algorithms by modifying them with preconditioners.
method Analyzes and compares computational costs of MCMC schemes with and without preconditioners.
result Establishes non-asymptotic guarantees for MCMC algorithms that learn and use preconditioners.
Develops algorithms for Bayesian inference with Plug & Play priors, ensuring convergence and well-posedness.
problem Bayesian imaging inverse problems with implicit priors defined by denoising algorithms.
method Introduces PnP-ULA and PnP-SGD algorithms for Monte Carlo sampling and MAP inference, proving convergence under realistic assumptions.
result Proves convergence of PnP-ULA and PnP-SGD algorithms for Bayesian inference with PnP priors, targeting a well-posed decision-theoretic model.
Paper proposes an algorithm for sampling from complex mixture distributions without requiring smoothness.
problem Sampling from a mixture of weakly smooth potentials.
method Unadjusted Langevin algorithm with Euler discretization for a mixture of weakly smooth distributions.
result Convergence in Kullback-Leibler divergence and L β L_β L β -Wasserstein metric with polynomial dependence on dimension. New method tunes SMC samplers efficiently without high costs.
problem Tuning SMC samplers with unadjusted kernels is challenging.
method Greedy Incremental Divergence Minimization (GIDM) for step size tuning.
result GIDM reduces KL divergence and tunes SMC samplers efficiently.