Langevin Monte Carlo for strongly log-concave distributions: Randomized midpoint revisited

Authors: Lu Yu, Avetik Karagulyan, Arnak S. Dalalyan

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we compare the performance of LMC, KLMC, RLMC, and RKLMC algorithms. We apply the four algorithms to the posterior density of penalized logistic regression, defined by π(ϑ) exp( f(ϑ)), with the potential function 2 ϑ 2 + 1 ndata i=1 log(1 + exp( yix i ϑ)) , where λ > 0 denotes the tuning parameter. The data {xi, yi}m i=1, composed of binary labels yi { 1, 1} and features xi Rp generated from xi,j iid N(0, 1), N(0, 5), and N(0, 10), corresponding to the plots from left to right, respectively. In our experiments, we have chosen λ = 1/100, p = 3 and ndata = 100. Figure 1 shows the W2-distance measured along the first dimension between the empirical distributions of the samples from the four algorithms and the target distribution4, with different choices of h. These numerical results confirm our theoretical results.
Researcher Affiliation Academia Lu Yu CREST, ENSAE, IP Paris lu.yu@ensae.fr Avetik Karagulyan KAUST avetik.karagulyan@kaust.edu.sa Arnak Dalalyan CREST, ENSAE, IP Paris arnak.dalalyan@ensae.fr
Pseudocode No The paper describes algorithms through mathematical equations and textual explanations, but it does not include any clearly labeled pseudocode blocks or algorithm figures.
Open Source Code No The paper does not contain an explicit statement about releasing open-source code for the described methodology, nor does it provide any links to a code repository.
Open Datasets No The paper describes generating synthetic data for its experiments: "The data {xi, yi}m i=1, composed of binary labels yi { 1, 1} and features xi Rp generated from xi,j iid N(0, 1), N(0, 5), and N(0, 10)". It does not reference a publicly available dataset with a citation, link, or repository.
Dataset Splits No The paper does not specify training, validation, or test dataset splits. It describes generating data for the numerical experiments but does not mention how this data was partitioned.
Hardware Specification No The paper does not provide any specific details about the hardware used for running the experiments, such as CPU or GPU models.
Software Dependencies No The paper does not provide specific software dependencies with version numbers needed to replicate the experiments.
Experiment Setup Yes In our experiments, we have chosen λ = 1/100, p = 3 and ndata = 100. ... Figure 1 shows the W2-distance measured along the first dimension between the empirical distributions of the samples from the four algorithms and the target distribution4, with different choices of h.