Communication-Efficient Distributed Stochastic AUC Maximization with Deep Neural Networks

Authors: Zhishuai Guo, Mingrui Liu, Zhuoning Yuan, Li Shen, Wei Liu, Tianbao Yang

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments on several benchmark datasets show the effectiveness of our algorithm and also confirm our theory.
Researcher Affiliation Collaboration 1Department of Computer Science, The University of Iowa, Iowa City, IA 52242, USA 2Tencent AI Lab, Shenzhen, China.
Pseudocode Yes Algorithm 1 Co DA; Algorithm 2 DSG
Open Source Code No The paper does not contain any explicit statements or links indicating that the source code for the described methodology is publicly available.
Open Datasets Yes We do experiments on 3 datasets: Cifar10, Cifar100 and Image Net.
Dataset Splits No The paper specifies training and testing data splits, but does not explicitly mention a separate validation set or split percentages for one.
Hardware Specification No The paper mentions 'a cluster of 4 computing nodes with each computer node having 4 GPUs', and states 'one machine corresponds to one GPU'. However, it does not specify the models of the GPUs (e.g., NVIDIA A100, Tesla V100) or any CPU details.
Software Dependencies No All algorithms are implemented by Py Torch (Paszke et al., 2019). The paper mentions PyTorch but does not specify a version number for it or any other software dependency.
Experiment Setup Yes For all algorithms, we set Ts = T03k, ηs = η0/3k. T0 and η0 are tuned for PPD-SG and set to the same for all other algorithms for fair comparison. T0 is tuned in [2000, 5000, 10000], and η0 is tuned in [0.1, 0.01, 0.001]. We fix the batch size for each GPU as 32.