Contrastively Disentangled Sequential Variational Autoencoder

Authors: Junwen Bai, Weiran Wang, Carla P. Gomes

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments show that C-DSVAE significantly outperforms the previous state-of-the-art methods on multiple metrics.
Researcher Affiliation Collaboration Junwen Bai Cornell University jb2467@cornell.edu Weiran Wang Google weiranwang@google.com Carla Gomes Cornell University gomes@cs.cornell.edu
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access to source code for the methodology described, nor does it explicitly state that the code is publicly available.
Open Datasets Yes Sprites [51] is a cartoon character video dataset. MUG [52] is a facial expression video dataset. SM-MNIST ([53], Stochastic Moving MNIST) is a dataset... TIMIT [54] is a corpus of read speech...
Dataset Splits Yes All datasets are separated into training, validation and testing splits following [46, 33, 18].
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory, or cloud instances) used for running the experiments.
Software Dependencies No The paper mentions software components like LSTMs and the Adam optimizer but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes As is commonly done in VAE learning [20], in (6) we add coefficients α to the KL terms, β to the contrastive terms, while the coefficient γ of I(s; z1:T ) is fixed to be 1. In our experiments, α is tuned over {0.6, 0.9, 1.0, 2.0} and β is tuned over {0.1, 0.2, 0.5, 0.7, 1.0, 2.0, 5.0}. We use the Adam optimizer [57] with the learning rate chosen from {0.0005, 0.001, 0.0015, 0.002} through grid search.