On Scalable and Efficient Computation of Large Scale Optimal Transport

Authors: Yujia Xie, Minshuo Chen, Haoming Jiang, Tuo Zhao, Hongyuan Zha

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical experiments on both synthetic and real datasets illustrate that SPOT is robust and has favorable convergence behavior.
Researcher Affiliation Academia Yujia Xie 1 Minshuo Chen 1 Haoming Jiang 1 Tuo Zhao 1 Hongyuan Zha 1 1Georgia Tech.
Pseudocode Yes Algorithm 1 Mini-batch Primal Dual Stochastic Gradient Algorithm for SPOT
Open Source Code No The paper does not include a statement that its own source code is being released or provide a link to its repository. It only links to the code of a comparison method (Deep JDOT).
Open Datasets Yes We evaluate DASPOT with the MNIST, MNISTM, USPS (Hull, 1994), and SVHN (Netzer et al., 2011) datasets.
Dataset Splits No The paper does not specify precise dataset split percentages or sample counts for training, validation, and testing. It refers to using datasets for tasks like 'Source Domain -> Target Domain' but not the specific partitioning methodology.
Hardware Specification Yes All experiments are implemented with Py Torch using one GTX1080Ti GPU and a Linux desktop computer with 32GB memory
Software Dependencies No The paper mentions 'Py Torch' but does not specify its version number or any other software dependencies with version details.
Experiment Setup Yes We adopt the Adam optimizer with configuration parameters 0.5 and 0.999 (Kingma & Ba, 2014). [...] We set da = 0 for the first 105 iteration to wait the generators to be well trained. Then we set da = 10 for the next 3 105 iteration. We take totally 4 105 iterations, and set the learning rate equal to 10 4 and batch size equal to 128 for all experiments.