Directed Chain Generative Adversarial Networks

Authors: Ming Min, Ruimeng Hu, Tomoyuki Ichiba

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The proposed DC-GANs are examined on four datasets, including two stochastic models from social sciences and computational neuroscience, and two real-world datasets on stock prices and energy consumption.
Researcher Affiliation Academia 1Department of Statistics and Applied Probability, University of California, Santa Barbara, CA 93106-3110, USA. 2Department of Mathematics, University of California, Santa Barbara, CA 93106-3080, USA.
Pseudocode Yes Algorithm 1 Generator in the Decorrelating and Branching
Open Source Code Yes The example implementation of DCGAN is available at https://github.com/mmin0/Directed_Chain_SDE.
Open Datasets Yes The third real-world data set of stock price time series was extracted from Yahoo Finance1, and the fourth real-world energy consumption data were obtained from Ireland s open data portal2. 1https://finance.yahoo.com/quote/GOOG?p=GOOG&.tsrc=fin-srch. 2https://data.gov.ie/dataset?theme=Energy.
Dataset Splits No We first generate the same amount of fake data paths as true data paths to avoid imbalance, and choose 80% from both real and fake data as training data, leaving the rest 20% as testing data.
Hardware Specification No The paper does not provide specific details about the hardware used for experiments, such as exact GPU or CPU models.
Software Dependencies No We remark that DC-GANs can be adapted to torchsde3 framework and use their adjoint method for back-propagation. 3See the Python package https://github.com/google-research/torchsde.
Experiment Setup Yes We use a two-layer LSTM classifier with channels/2 as the size of the hidden state, where channels is the dimension of generated and real series. We will minimize the cross-entropy loss, and the optimization is done by Adam optimizer with a learning rate of 0.001 for 5000 iterations.