A Faster Decentralized Algorithm for Nonconvex Minimax Problems
Authors: Wenhan Xian, Feihu Huang, Yanfu Zhang, Heng Huang
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experiments on decentralized settings show the superior performance of our new algorithm. |
| Researcher Affiliation | Academia | Wenhan Xian, Feihu Huang, Yanfu Zhang, Heng Huang Electrical and Computer Engineering, University of Pittsburgh, Pittsburgh, PA 15213 wex37@pitt.edu, huangfeihu2018@gmail.com, yaz91@pitt.edu, heng.huang@pitt.edu |
| Pseudocode | Yes | Algorithm 1 DM-HSGD |
| Open Source Code | Yes | 1https://github.com/Trashzz Z/DM-HSGD |
| Open Datasets | Yes | We conduct our experiment on six real-world training datasets a9a", covtype", ijcnn1", phishing", rcv1" and w8a", which can be downloaded from LIBSVM2 repository. 2https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets |
| Dataset Splits | No | The paper specifies using 'mini-batch size is set to 20' for experiments but does not explicitly describe how the datasets were split into training, validation, or test sets, nor does it refer to predefined splits for these purposes. |
| Hardware Specification | Yes | We implement our code on an MPI cluster where each node is equipped with 12-core Intel Xeon E5-2620 v3 2.40 GHz processor. |
| Software Dependencies | No | The paper does not provide specific version numbers for any software dependencies used in the experiments (e.g., Python, deep learning frameworks like PyTorch or TensorFlow, or other libraries). |
| Experiment Setup | Yes | For each algorithm, we grid search the learning rates ηx and ηy from {0.1, 0.01, 0.001, 0.0001}. The mini-batch size is set to 20. The number of iterations in the nested loop for double-loop algorithms is set to K = 5. For DM-HSGD, we set the batch size of the first iteration to b0 = 10000. βx and βy are set to 0.01. For SREDA, we set ϵ = 0.1 in the factor ϵ vt , period q = 50 and large batch size S1 = 1000. |