Continuous-Time Flows for Efficient Inference and Density Estimation

Authors: Changyou Chen, Chunyuan Li, Liqun Chen, Wenlin Wang, Yunchen Pu, Lawrence Carin Duke

ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on various tasks demonstrate promising performance of the proposed CTF framework, compared to related techniques.
Researcher Affiliation Academia 1SUNY at Buffalo 2Duke University.
Pseudocode Yes The algorithm is presented in Algorithm 1 in Section E of the SM.
Open Source Code No Some experiments are based on the excellent code for Stein GANk (Wang & Liu, 2017), where their default parameter setting are adopted. khttps://github.com/Dart ML/Stein GAN. There is no explicit statement or link provided for the authors' own code for the proposed CTF framework.
Open Datasets Yes We test Mac GAN on three datasets: MNIST, CIFAR-10 and Celab A.
Dataset Splits No The paper mentions 'training epochs' and 'testing ELBOs' but does not explicitly provide details about validation dataset splits (e.g., percentages or counts) or a distinct validation set being used.
Hardware Specification No The paper does not explicitly describe the hardware used to run its experiments, such as specific GPU or CPU models.
Software Dependencies No The paper mentions 'DCGAN architecture' but does not provide specific version numbers for any software dependencies, programming languages, or libraries used in the experiments.
Experiment Setup Yes We define the inference network as a deep neural network with two fully connected layers of size 300 with softplus activation functions. The generator Gφ is defined as a 3-layer CNN with the Re LU activation function (except for the top layer which uses tanh as the activation function, see SM G for details). Following (Wang & Liu, 2017), the stepsizes are set to (me e) lr me 50 , where e indexes the epoch, me is the total number of epochs, lr = 1e-4 when updating , and lr = 1e-3 when updating φ. The stepsize in L1 is set to 1e-3.