Finding Second-Order Stationary Points in Nonconvex-Strongly-Concave Minimax Optimization

Authors: Luo Luo, Yujun Li, Cheng Chen

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we conduct empirical studies for our methods against the classical GDA algorithm [20] on both synthetic problem and real-world application.
Researcher Affiliation Collaboration Luo Luo School of Data Science Fudan University luoluo@fudan.edu.cn Yujun Li Noah s Ark Lab Huawei Technologies Co., Ltd. liyujun9@huawei.com Cheng Chen School of Physical and Mathematical Sciences Nanyang Technological University cheng.chen@ntu.edu.sg
Pseudocode Yes Algorithm 1 AGD(h, y0, K, , ), Algorithm 2 Minimax Cubic-Newton (MCN), Algorithm 3 Inexact Minimax Cubic-Newton, Algorithm 4 Cubic-Solver, Algorithm 5 Final-Cubic-Solver
Open Source Code Yes Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes]
Open Datasets Yes We compare IMCN with GDA on the domain adaptation problem between two different datasets: MNIST [19] and MNIST-m [9].
Dataset Splits No The paper mentions using MNIST and MNIST-m datasets but does not explicitly state the training, validation, or test splits within the provided text.
Hardware Specification No The paper states in its checklist that it includes the type of resources used, but these specific details (e.g., GPU/CPU models) are not present in the provided main paper text.
Software Dependencies No The paper does not provide specific software dependencies with version numbers in the provided text.
Experiment Setup Yes The learning rate of GDA and AGD step in MCN is selected from - c 10 i : c 2 {1, 5}, i 2 {1, 2, 3}. For MCN method, we choose M = 10.