Exemplar VAE: Linking Generative Models, Nearest Neighbor Retrieval, and Data Augmentation

Authors: Sajad Norouzi, David J. Fleet, Mohammad Norouzi

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments demonstrate the effectiveness of Exemplar VAEs on density estimation and representation learning. Importantly, generative data augmentation using Exemplar VAEs on permutation invariant MNIST and Fashion MNIST reduces classification error from 1.17% to 0.69% and from 8.56% to 8.16%.
Researcher Affiliation Collaboration 1University of Toronto, 2Vector Institute, 3Google Research
Pseudocode No The paper describes the generative process of the Exemplar VAE in three steps in Section 3, but this is presented as descriptive text rather than a structured pseudocode or algorithm block.
Open Source Code Yes Code is available at https://github.com/sajadn/Exemplar-VAE.
Open Datasets Yes We use four datasets, namely, MNIST, Fashion-MNIST, Omniglot, and Celeb A
Dataset Splits Yes Figure 2: Training and validation ELBO on Dynamic MNIST for Exemplar VAE with and without LOO.
Hardware Specification No The paper mentions training 'convolutional models' and 'convolutional architectures' but does not specify any particular GPU or CPU models, memory sizes, or other specific hardware components used for the experiments.
Software Dependencies No The paper mentions using 'gradient normalized Adam [31, 58]' as an optimizer, but it does not specify any software versions for programming languages (e.g., Python), deep learning frameworks (e.g., PyTorch, TensorFlow), or other libraries.
Experiment Setup Yes We use gradient normalized Adam [31, 58] with learning rate 5e-4 and linear KL annealing for 100 epochs.