Finite-Time Analysis of Projected Langevin Monte Carlo

Authors: Sebastien Bubeck, Ronen Eldan, Joseph Lehec

NeurIPS 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We ran the volume algorithm with both H&R and LMC on the following set of convex bodies: K = [ 1, 1]n (referred to as the Box ) and K = [ 1, 1]n n 2 Bn (referred to as the Box and Ball ), where n = 10 k, k = 1, . . . , 10. The computed volume (normalized by 2n for the Box and by 0.2 2n for the Box and Ball ) as well as the clock time (in seconds) to terminate are reported in the figure above. From these experiments it seems that LMC and H&R roughly compute similar values for the volume (with H&R being slightly more accurate), and LMC is almost always a bit faster.
Researcher Affiliation Collaboration S ebastien Bubeck Microsoft Research sebubeck@microsoft.com Ronen Eldan Weizmann Institute roneneldan@gmail.com Joseph Lehec Universit e Paris-Dauphine lehec@ceremade.dauphine.fr
Pseudocode No The paper does not contain any pseudocode or clearly labeled algorithm blocks. The algorithm (1) is presented as a mathematical formula.
Open Source Code No The paper does not provide any explicit statements about releasing source code or links to a code repository.
Open Datasets No The 'datasets' used are geometric shapes (Box and Box and Ball) defined by mathematical expressions, not publicly available datasets with specific access information. The paper does not provide a link, DOI, or formal citation for accessing these geometric definitions as data.
Dataset Splits No The paper does not specify training, validation, or test dataset splits. The experiments involve computing the volume of geometric bodies, not training models with data splits.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, or memory) used for running the experiments.
Software Dependencies No The paper mentions 'Cousins and Vempala provide a Matlab implementation' but does not specify a version number for Matlab or any other software dependencies with versions.
Experiment Setup Yes We implemented the same procedure with LMC instead of H&R, and we choose the step-size η = 1/(βn2), where β is the smoothness parameter of the underlying log-concave distribution (in particular here β = 1/σ2 ℓ).