High Probability Generalization Bounds with Fast Rates for Minimax Problems
Authors: Shaojie Li, Yong Liu
ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we report preliminary experimental results to verify our theoretical results by performing numerical experiments on the simulated data. |
| Researcher Affiliation | Academia | Shaojie Li1,2, Yong Liu1,2, 1Gaoling School of Artiļ¬cial Intelligence, Renmin University of China, Beijing, China 2Beijing Key Laboratory of Big Data Management and Analysis Methods, Beijing, China |
| Pseudocode | No | The paper describes algorithms (GDA, SGDA, PPM, EG, OGDA) using mathematical update equations, but it does not include any structured pseudocode blocks or algorithms explicitly labeled as such. |
| Open Source Code | No | The paper does not contain any statement about releasing source code or links to a code repository for the described methodology. |
| Open Datasets | No | We consider an isotropic Gaussian data vector Z N(0, Id d) with zero mean and identity covariance. We will draw n independent samples from the underlying Gaussian distribution to form a training dataset S = {z1, ..., zn}. |
| Dataset Splits | No | The paper uses simulated data and evaluates the generalization error with respect to the number of samples, but it does not explicitly define training, validation, or test dataset splits. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU/CPU models, memory) used to run the experiments. |
| Software Dependencies | No | The paper does not specify any software names with version numbers that would be needed to replicate the experiments. |
| Experiment Setup | Yes | For GDA and SGDA, we consider the stepsize parameter as 1/t. We iterate GDA with n2 times and SGDA with n4 times. The generalization error of GDA and SGDA with different sizes of training data are reported in Figure 1. And for EG and OGDA, we select the stepsize parameter as 0.003. We run EG and OGDA n2 times. |