Accelerated Mini-batch Randomized Block Coordinate Descent Method

Authors: Tuo Zhao, Mo Yu, Yiming Wang, Raman Arora, Han Liu

NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our numerical experiments shows that the MRBCD method naturally exploits the sparsity structure and achieves better computational performance than existing methods. 5 Numerical Simulations 6 Real Data Example
Researcher Affiliation Academia Johns Hopkins University Harbin Institute of Technology Princeton University {tour,myu25,freewym,arora}@jhu.edu,hanliu@princeton.edu
Pseudocode Yes Algorithm 1 Mini-batch Randomized Block Coordinate Descent Method-I: A Naive Implementation. Algorithm 2 Mini-batch Randomized Block Coordinate Descent Method-II: MRBCD + Variance Reduction. Algorithm 3 Mini-batch Randomized Block Coordinate Descent Method-III: MRBCD with Variance Reduction and Active Set.
Open Source Code No The paper does not provide any link or explicit statement about the availability of open-source code for the described methodology.
Open Datasets Yes The second sparse learning problem is the elastic-net regularized logistic regression, which solves fi( ) + λ1|| ||1 with fi = log(1 + exp( yix T)). We adopt the rcv1 dataset with n = 20242 and d = 47236.
Dataset Splits No The paper mentions data generation for synthetic data and dataset sizes for real data (rcv1) but does not provide explicit training, validation, or test dataset splits. It does not mention any specific percentages, counts, or references to standard splits for these datasets.
Hardware Specification No The paper does not specify any hardware details (e.g., CPU, GPU models, memory, or cloud instances) used for running the experiments.
Software Dependencies No The paper does not mention any specific software dependencies or their version numbers (e.g., Python, PyTorch, TensorFlow, or specific solvers).
Experiment Setup Yes We set n = 2000 and d = 1000... We choose λ = log d/n... We set k = 100. All blocks are of the same size (10 coordinates). For BPG, the step size is 1/T... For SPVRG, we choose m = n, and the step size is 1/(4T). For MRBCD-II, we choose m = n, and the step size is 1/(4L). Note that the step size and number of iterations m within each inner loop for MRBCD-II and SPVRG are tuned over a refined grid such that the best computational performance is obtained.