Variance reduction for Random Coordinate Descent-Langevin Monte Carlo

Authors: ZHIYAN DING, Qin Li

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate numerical evidence in Section 6. Proofs are rather technical and are all left to appendices.
Researcher Affiliation Academia Zhiyan Ding Department of Mathematics University of Wisconsin-Madison Madison, WI 53706 zding49@math.wisc.edu Qin Li Department of Mathematics University of Wisconsin-Madison Madison, WI 53706 qinli@math.wisc.edu
Pseudocode Yes Algorithm 1 Randomized Coordinate Averaging Decent O/U-LMC (RCAD-O/U-LMC)
Open Source Code No The paper does not provide an explicit statement or link to open-source code for the methodology described.
Open Datasets No The paper describes synthetic target distributions (e.g., N(0, Id) and f(x) = (x1 − 1)^2 + Σ(x_i)^2) and initial distributions, but does not refer to a publicly available dataset with concrete access information (link, DOI, citation).
Dataset Splits No The paper mentions running simulations with N = 5 * 10^5 particles and discusses initial distributions, but does not specify dataset splits (e.g., training, validation, test percentages or counts) as it deals with sampling from a distribution rather than using a fixed dataset.
Hardware Specification No The paper does not specify any hardware details (e.g., CPU, GPU models, memory) used for running the experiments.
Software Dependencies No The paper does not specify any software dependencies with version numbers.
Experiment Setup Yes In the first example, our target distribution is N(0, Id) with d = 1000, and in the second example we use f(x) = (x1 − 1)^2 + Σdi=2 x2i. The initial distributions, for the overdamped and underdamped situations respectively, are N(0.5, Id) and N(0.5, I2d) in both exampes. We run both RCD-O/U-LMC and RCAD-O/U-LMC using N = 5 * 10^5 particles and test MSE error with φ(x) = |x1|^2 in both examples. In all the computation, M is big enough. The improvement of adding variance reduction technique is obvious in both examples.