Zeroth-Order Methods for Constrained Nonconvex Nonsmooth Stochastic Optimization
Authors: Zhuanghua Liu, Cheng Chen, Luo Luo, Bryan Kian Hsiang Low
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we conduct numerical experiments that demonstrate the effectiveness of our algorithms. |
| Researcher Affiliation | Academia | 1Department of Computer Science, National University of Singapore, Singapore, Singapore 2CNRS@CREATE LTD, 1 Create Way, #08-01 CREATE Tower, Singapore 138602 3Shanghai Key Laboratory of Trustworthy Computing, East China Normal University, Shanghai, China 4School of Data Science, Fudan University, Shanghai, China 5Shanghai Key Laboratory for Contemporary Applied Mathematics. Correspondence to: Luo Luo <luoluo@fudan.edu.cn>. |
| Pseudocode | Yes | Algorithm 1 vt = MB-SGrad(xt), Algorithm 2 vt = VR-SGrad(xt, xt 1, t), Algorithm 3 ZOSPGD Method, Algorithm 4 ZOSFW Method, Algorithm 5 Two-Phase Zeroth-Order Stochastic Projected Gradient Descent Method, Algorithm 6 Two-Phase Zeroth-Order Stochastic Frank Wolfe Method |
| Open Source Code | No | The paper does not provide an explicit statement or link for open-source code for the described methodology. |
| Open Datasets | Yes | For the real-world dataset, we validate our methods on the Movie Lens 1M 2 dataset. The dataset is a sparse movie rating matrix Y with 6040 users and 3952 movies. Each rating of the matrix Y is an integer ranging from 1 to 5. We set the parameter B = 7000. 2https://grouplens.org/datasets/movielens/1m/ |
| Dataset Splits | No | The paper does not explicitly provide specific train/validation/test dataset splits with percentages, sample counts, or references to predefined splits. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU or CPU models used for running experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers. |
| Experiment Setup | Yes | For all the experiments, we set the parameter σ = 1. For the synthetic dataset... we choose the parameter B = 2000 while for d = 5000, we choose B = 4000. ... we choose the minibatch size b = 100, 000 for both MB-ZOSPGD and MB-ZOSFW methods. We set b1 = 100, 000, b2 = 10, 000 and q = b1/b2 for VR-ZOSPGD and VR-ZOSFW methods. The number of iterations T is set to be 300 for all algorithms. The step size is tuned from the set {0.1, 0.03, . . . , 3 10 7, 1 10 7} for each algorithm. |