A Simple yet Universal Strategy for Online Convex Optimization
Authors: Lijun Zhang, Guanghui Wang, Jinfeng Yi, Tianbao Yang
ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct preliminary experiments to evaluate the proposed USC, and the detail is provided in Appendix C. We investigate both strongly convex functions and general convex functions, and present the results in Fig. 1. |
| Researcher Affiliation | Collaboration | 1National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, China 2Peng Cheng Laboratory, Shenzhen, China 3Frontis.AI, Beijing, China 4Department of Computer Science, The University of Iowa, Iowa City, USA. |
| Pseudocode | Yes | Our universal strategy for online convex optimization (USC) is summarized in Algorithm 1. [...] Algorithm 1 A Universal Strategy for Online Convex Optimization (USC) |
| Open Source Code | No | The paper does not provide any statement or link indicating that the source code for the described methodology is publicly available. |
| Open Datasets | No | We conduct both experiments on the a9a dataset (Chang & Lin, 2011). While a citation is provided, there is no explicit link, DOI, or repository information for accessing the dataset. |
| Dataset Splits | No | The paper mentions using the a9a dataset and sampling data points, but it does not specify explicit training, validation, or test dataset splits. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU/GPU models, memory specifications) used for running the experiments. |
| Software Dependencies | No | The paper mentions various algorithms (OGD, Adam, ONS, etc.) but does not list specific software libraries or their version numbers used in the implementation. |
| Experiment Setup | Yes | We configure λ = 0.02, the diameter of the decision set D = 20, and the time horizon T = 10000. [...] Acon includes OGD with step size ηt = G D t, and Adam with hype-parameters β1 = 0.9, β2 = 0.999, and different step sizes in range {1, 10 1, . . . , 10 4}. [...] Astr consists of SC-OGD with step size ηt = 1 λt, and SAdam with hype-parameters β1 = 0.9, β2 = 1 0.9 t , and step size α = 0.1 λ. |