Chain of Log-Concave Markov Chains

Authors: Saeed Saremi, Ji Won Park, Francis Bach

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We study our sampling algorithm quantitatively using the 2-Wasserstein metric and compare it with various Langevin MCMC algorithms. We also report a remarkable capacity of our algorithm to tunnel between modes of a distribution.
Researcher Affiliation Collaboration Saeed Saremi1, Ji Won Park1, Francis Bach2 1Frontier Research, Prescient Design, Genentech, South San Francisco, CA 2Inria, Ecole Normale Supérieure, Université PSL, Paris, France
Pseudocode Yes Algorithm 1: Sequential multimeasurement walk-jump sampling referred to by SMS.
Open Source Code No The paper does not provide any explicit statements about code availability or links to source code repositories for the described methodology.
Open Datasets No The paper uses mathematically defined test densities (e.g., Elliptical Gaussian, Mixture of Gaussians) which do not require external public datasets or specific access information.
Dataset Splits No The paper conducts experiments on mathematically defined test densities and does not involve traditional dataset splits for training, validation, or testing.
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments.
Software Dependencies No The paper mentions specific MCMC algorithms and integration schemes (e.g., 'Sachs et al. (2017)'), but does not list specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes The hyperparameters were tuned on a log-spaced grid. We searched the step size δ over {0.03, 0.1, 0.3, 1.0}, the effective friction γδ over {0.0625, 0.125, 0.25, 0.5, 1.0}, per-t MCMC iterations nt over {1, 4, 16}, and the Lipschitz parameter over {1/σ2, 1.0}.