Abstract:Stochastic Gradient Langevin Dynamics (SGLD) have been widely used for Bayesian sampling from certain probability distributions, incorporating derivatives of the log-posterior. With the derivative evaluation of the log-posterior distribution, SGLD methods generate samples from the distribution through performing as a thermostats dynamics that traverses over gradient flows of the log-posterior with certainly controllable perturbation. Even when the density is not known, existing solutions still can first learn the kernel density models from the given datasets, then produce new samples using the SGLD over the kernel density derivatives. In this work, instead of exploring new samples from kernel spaces, a novel SGLD sampler, namely, Randomized Measurement Langevin Dynamics (RMLD) is proposed to sample the high-dimensional sparse representations from the spectral domain of a given dataset. Specifically, given a random measurement matrix for sparse coding, RMLD first derives a novel likelihood evaluator of the probability distribution from the loss function of LASSO, then samples from the high-dimensional distribution using stochastic Langevin dynamics with derivatives of the logarithm likelihood and Metropolis–Hastings sampling. In addition, new samples in low-dimensional measuring spaces can be regenerated using the sampled high-dimensional vectors and the measurement matrix. The algorithm analysis shows that RMLD indeed projects a given dataset into a high-dimensional Gaussian distribution with Laplacian prior, then draw new sparse representation from the dataset through performing SGLD over the distribution. Extensive experiments have been conducted to evaluate the proposed algorithm using real-world datasets. The performance comparisons on three real-world applications demonstrate the superior performance of RMLD beyond baseline methods.

Sparse Representation of Gaussian Molecular Surface

Sparse Representation Based on Projection Method in Online Least Squares Support Vector Machines

Sparse RBF Surface Representations.

Sparse Representation for Three-Dimensional Number Ball Recognition

Description Of Complex Optical Surface Based On Sparse Radial Basis Function Approximation With Spatially Variable Shape Parameters

Representation of Complex Optical Surfaces with Adaptive Radial Basis Functions

Feature Sensitive Re-Sampling of Point Set Surfaces with Gaussian Spheres

Surface Approximation Via Sparse Representation and Parameterization Optimization

Shape Sparse Correspondence for Nearly-Isometric Meshes Based on Green's Function Representation

A Molecular Stereostructure Descriptor Based on Spherical Projection

Efficient non-parametric fitting of potential energy surfaces for polyatomic molecules with Gaussian processes

Sampling Sparse Representations with Randomized Measurement Langevin Dynamics

Sparse Gaussian processes for solving nonlinear PDEs

Sparse low rank approximation of potential energy surfaces with applications in estimation of anharmonic zero point energies and frequencies

Robust Gaussian Process Regression method for efficient reaction pathway optimization: application to surface processes

A New Split Algorithm for 3D Gaussian Splatting

3DGSR: Implicit Surface Reconstruction with 3D Gaussian Splatting

Gaussian Moments as Physically Inspired Molecular Descriptors for Accurate and Scalable Machine Learning Potentials

Riemannian Geometry and Molecular Surfaces I: Spectrum of the Laplacian

Learning Harmonic Molecular Representations on Riemannian Manifold

Adaptive Algorithm for Electronic Structure Calculations Using Reduction of Gaussian Mixtures