ADMM and Accelerated ADMM as Continuous Dynamical Systems

Authors: Guilherme Franca, Daniel Robinson, Rene Vidal

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

Reproducibility Variable Result LLM Response
Research Type Experimental We numerically verify that the differential equations (10) and (21) accurately model ADMM and A-ADMM, respectively, when ρ is large as needed to derive the continuous limit. The numerical integration of the first-order system (10) is straightforward; we use a 4th order Runge-Kutta method (an explicit Euler method could also be employed). The numerical integration of (21) is more challenging due to strong oscillations. To obtain a faithful discretization of the continuous dynamical system (21), i.e., one that preserves its properties, a standard approach is to use a Hamiltonian symplectic integrator, which is designed to preserve the phase-space volume. A simple example is provided in Figure 1, which illustrate our theoretical results.
Researcher Affiliation Academia 1Mathematical Institute for Data Science, Johns Hopkins University, Baltimore MD 21218, USA.
Pseudocode No The paper presents mathematical updates for algorithms (e.g., equations 9 and 19) and numerical integration schemes (equation 59), but these are presented as mathematical expressions rather than structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any statement about releasing source code for the described methodology.
Open Datasets No The paper uses synthetic data described as "a random matrix with 40 zero eigenvalues and the remaining ones are uniformly distributed on [0, 10], and A is a full column random matrix with condition number 100" in Figure 1, which is not a publicly available dataset nor is any access information provided.
Dataset Splits No The paper does not specify training, validation, or test dataset splits. The numerical example in Figure 1 uses a single initial condition for a simulation rather than a dataset with splits.
Hardware Specification No The paper does not provide specific details about the hardware used to run the numerical experiments.
Software Dependencies No The paper mentions numerical methods like "4th order Runge-Kutta method" and "symplectic Euler method" but does not specify any software names or version numbers (e.g., Python, PyTorch, MATLAB, specific solvers) used for implementation.
Experiment Setup Yes We choose r = 10 and ρ = 50. The initial conditions are X(0) = x0 = 5(1, 1, . . . , 1)T and X(0) = 0.