Elastic-InfoGAN: Unsupervised Disentangled Representation Learning in Class-Imbalanced Data

Authors: Utkarsh Ojha, Krishna Kumar Singh, Cho-Jui Hsieh, Yong Jae Lee

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on both artificial (MNIST, 3D cars, 3D chairs, Shape Net) and real-world (You Tube-Faces) imbalanced datasets demonstrate the effectiveness of our method in disentangling object identity as a latent factor of variation.
Researcher Affiliation Collaboration Utkarsh Ojha1 Krishna Kumar Singh1,2 Cho-Jui Hsieh3 Yong Jae Lee1 1UC Davis 2Adobe Research 3UCLA
Pseudocode No The paper includes mathematical formulations and block diagrams (e.g., Figure 2) to illustrate the model, but it does not contain any sections explicitly labeled 'Pseudocode' or 'Algorithm', nor are there any structured code-like procedural steps.
Open Source Code Yes utkarshojha.github.io/elastic-infogan/
Open Datasets Yes Datasets (1) MNIST [34]... (2) 3D Cars [17]... (3) 3D Chairs [2]... (4) Shape Net... (5) You Tube-Faces [51]...
Dataset Splits Yes We train the classifier by creating a 80/20 train/val split on a per class basis.
Hardware Specification No The paper does not provide any specific details regarding the hardware specifications (e.g., GPU model, CPU, memory) used for conducting the experiments.
Software Dependencies No The paper discusses the use of Gumbel-Softmax as a technique and implies common deep learning frameworks, but it does not explicitly list any software dependencies with specific version numbers (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup No The paper mentions parameters like softmax temperature (τ) and loss weights (λ1, λ2) in the equations and discussion. However, it lacks specific numerical values for common experimental setup details such as learning rates, batch sizes, optimizers, or the number of training epochs in the main text.