NysADMM: faster composite convex optimization via low-rank approximation

Authors: Shipu Zhao, Zachary Frangella, Madeleine Udell

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical experiments on real-world datasets demonstrate that Nys ADMM can solve important applications, such as the lasso, logistic regression, and support vector machines, in half the time (or less) required by standard solvers.
Researcher Affiliation Academia 1Cornell University, Ithaca, NY, USA. 2Stanford University, Stanford, CA, USA.
Pseudocode Yes Algorithm 1 ADMM input feature matrix A, response b, loss function ℓ, regularization g and h, stepsize ρ repeat... Algorithm 4 Randomized Nystr om Approximation... Algorithm 7 Ada Nys ADMM
Open Source Code No The paper does not contain an explicit statement about releasing source code for the described methodology or provide a link to a code repository.
Open Datasets Yes These experiments use datasets with n > 10, 000 or d > 10, 000 from LIBSVM (Chang & Lin, 2011), UCI (Dua & Graff, 2017), and Open ML (Vanschoren et al., 2013), with statistics summarized in Table 2.
Dataset Splits No The paper mentions using datasets for experiments and refers to 'E2006.train' in a dataset name, but it does not specify explicit train/validation/test dataset splits, percentages, or sample counts for reproducibility.
Hardware Specification Yes We run all experiments on a server with 128 Intel Xeon E74850 v4 2.10GHz CPU cores and 1056GB.
Software Dependencies No The paper mentions using 'SAGA algorithm...implemented in sklearn' and 'LIBSVM solver (Chang & Lin, 2011)', but it does not provide specific version numbers for these software components.
Experiment Setup Yes The tolerance of Nys ADMM at each iteration is chosen as the geometric mean εk+1 = q rkprk d of the ADMM primal residual rp and dual residual rd at the previous iteration... We choose a sketch size s = 50 to compute the Nystr om approximation throughout our experiments.