Gradient-Free Method for Heavily Constrained Nonconvex Optimization
Authors: Wanli Shi, Hongchang Gao, Bin Gu
ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results in two applications demonstrate the superiority of our method in terms of training time and accuracy compared with other ZO methods for the constrained problem. |
| Researcher Affiliation | Academia | 1Nanjing University of Information Science and Technology, Jiangsu, China 2MBZUAI, Abu Dhabi, UAE 3Department of Computer and Information Sciences, Temple University, PA, USA. |
| Pseudocode | Yes | Algorithm 1 Doubly Stochastic Zeroth-order Gradient (DSZOG). |
| Open Source Code | No | The paper does not contain any explicit statements or links indicating that the source code for the methodology is openly available. |
| Open Datasets | No | We summarized the datasets used in this application in Table 4. We generate 4 datasets with 2000 samples in this task and summarize them in Table 5. The paper mentions dataset names 'a9a', 'w8a', 'gen', 'svm', 'D1', 'D2', 'D3', 'D4' but does not provide explicit citations, links, or repository information for accessing them. |
| Dataset Splits | Yes | We randomly sample 1000 data samples from the original datasets, and then divide all the datasets into 3 parts, i.e., 50% for training, 30% for testing and 20% for validation. |
| Hardware Specification | Yes | We run all the methods 10 times on a 3990x workstation. |
| Software Dependencies | No | The paper does not provide specific ancillary software details with version numbers. |
| Experiment Setup | Yes | We fix the batch size of data sample at 128 for all the methods and |M2| = |M3| = 128. The learning rates of all the methods are chosen from {0.01, 0.001, 0.0001}. In our methods, the penalty parameter β is chosen from {0.1, 1, 10}, a and b are chosen from {0.1, 0.5, 0.9} on the validation sets. |