Optimal Stochastic and Online Learning with Individual Iterates
Authors: Yunwen Lei, Peng Yang, Ke Tang, Ding-Xuan Zhou
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We report experimental comparisons with several baseline methods to show the effectiveness of our method in achieving a fast training speed as well as in outputting sparse solutions. In this section, we justify the effectiveness of our algorithm by presenting experimental comparisons with the following averaging strategies: WEI-AVE [20], UNI-AVE, LAST, SUFFIX [31] and RAND (outputting a random iterate chosen from the uniform distribution over the last half of iterates). |
| Researcher Affiliation | Academia | Yunwen Lei1,2 Peng Yang1 Ke Tang1 Ding-Xuan Zhou3 1University Key Laboratory of Evolving Intelligent Systems of Guangdong Province, Department of Computer Science and Engineering, Southern University of Science and Technology, Shenzhen 518055, China 2Department of Computer Science, Technical University of Kaiserslautern, Kaiserslautern 67653, Germany 3School of Data Science and Department of Mathematics, City University of Hong Kong, Kowloon, Hong Kong, China {leiyw, yangp, tangk3}@sustech.edu.cn mazhou@cityu.edu.cn |
| Pseudocode | Yes | Algorithm 1: SCMDI Input: {ηt}t, σφ, w1 and T. Output: an approximate solution of (2.1) ... and Algorithm 2: OCMDI Input: {ηt}t, σφ and w1. Output: an approximate solution of (2.1) ... |
| Open Source Code | No | The paper does not contain any explicit statement about releasing source code for the described methodology or a link to a code repository. |
| Open Datasets | Yes | We use 16 real-world datasets whose information is summarized in Table C.1 in Appendix D.13. |
| Dataset Splits | Yes | For each dataset, we use 80 percents of the data for training and reserve the remaining 20 percents for testing. We consider step sizes of the form ηt = µ/(λt) and tune the parameter µ in the set 2{ 12, 11,...,4} by 10-fold cross validation. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments. It only mentions running 'experiments' without specifying the computational resources. |
| Software Dependencies | No | The paper mentions software like 'AIR toolbox' and 'MATLAB package' but does not provide specific version numbers for any software dependencies used in their experiments. For example, 'We use AIR toolbox [12] to create a CT-measurement matrix A Rn d, an output vector y Rn and a N N sparse image encoded by a vector w Rd with d = N 2.' |
| Experiment Setup | Yes | We consider step sizes of the form ηt = µ/(λt) and tune the parameter µ in the set 2{ 12, 11,...,4} by 10-fold cross validation. We randomly choose w1 from the uniform distribution in [0, 1]d and set λ = 1, ϵ = 10 8 as suggested in [5]. |