The Physical Systems Behind Optimization Algorithms

Authors: Lin Yang, Raman Arora, Vladimir braverman, Tuo Zhao

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical We present an illustration of our theoretical analysis in Figure 2. We consider a strongly convex quadratic program f(x) = 1/2xHx, where H = [300 1; 1 50]. Obviously, f(x) is strongly convex and x = [0, 0] is the minimizer. We choose η = 10^-4 for VGD and NAG, and η = 2*10^-4 for RCGD and ARCG. The trajectories of VGD and NAG are obtained by the default method for solving ODE in MATLAB.
Researcher Affiliation Academia Lin F. Yang Princeton University lin.yang@princeton.edu; Raman Arora, Johns Hopkins University arora@cs.jhu.edu; Vladimir Braverman Johns Hopkins University vova@cs.jhu.edu; Tuo Zhao Georgia Institute of Technology tourzhao@gatech.edu
Pseudocode No The paper describes algorithms such as VGD, NAG, RCGD, ARCG, and Newton's method in Appendix A, but it does so using text and mathematical equations, not structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any concrete access information for open-source code for the methodology described, such as a repository link, or explicit statements about code availability in supplementary materials.
Open Datasets No The paper uses a synthetic quadratic program for numerical simulations, not a publicly available dataset. No concrete access information for any dataset is provided.
Dataset Splits No The paper does not specify dataset split information (e.g., training, validation, test splits) as it focuses on theoretical analysis and uses a synthetic mathematical function for illustration rather than a traditional dataset.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, or memory specifications) used for running the numerical simulations.
Software Dependencies No The paper mentions 'MATLAB' for numerical simulations but does not provide specific version numbers for MATLAB or any other software dependencies.
Experiment Setup Yes We consider a strongly convex quadratic program f(x) = 1/2xHx, where H = [300 1; 1 50]. ... We choose η = 10^-4 for VGD and NAG, and η = 2*10^-4 for RCGD and ARCG. The trajectories of VGD and NAG are obtained by the default method for solving ODE in MATLAB.