AdaVAE: Bayesian Structural Adaptation for Variational Autoencoders

Authors: Paribesh Regmi, Rui Li

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments show that the inference framework effectively prevents overfitting in both shallow and deep VAE models, yielding state-of-the-art performance.
Researcher Affiliation Academia Paribesh Regmi Rui Li Rochester Institute of Technology {pr8537, rxlics}@rit.edu
Pseudocode No The paper describes the method mathematically but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes Implementation details are in the Appendix. Codes are provided.
Open Datasets Yes We assess our estimator using three benchmark datasets: MNIST [47], Omniglot [48], and Caltech101 Silhouettes [49].
Dataset Splits No The paper mentions using benchmark datasets but does not provide specific details on the train/validation/test splits, such as percentages, sample counts, or explicit standard split references for reproduction.
Hardware Specification No The paper states: 'We acknowledge Research Computing at the Rochester Institute of Technology [57] for providing computational resources.' but does not provide specific hardware details such as GPU/CPU models, memory, or processor types.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies (e.g., programming languages, libraries, or frameworks) used in the experiments.
Experiment Setup Yes To investigate how network structures evolve during training epochs, we set the truncation level T = 25 on MLP backbone nets with tanh non-linearities. ... We examine four settings of the VAE structure sample size S = {1, 2, 4, 8}, along with latent variable sample sizes (M, K) = {(8, 8), (4, 16)} as in [18]. The truncation level is T = 25 with a maximum width O = 200. The distribution over the output from the decoding networks is factorized Bernoulli. ... All methods share the same maximum width of O = 200 and a latent variable dimensionality of 50.