Disentangling Factors of Variations Using Few Labels
Authors: Francesco Locatello, Michael Tschannen, Stefan Bauer, Gunnar Rätsch, Bernhard Schölkopf, Olivier Bachem
ICLR 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this paper, we investigate the impact of such supervision on state-of-the-art disentanglement methods and perform a large scale study, training over 52 000 models under well-defined and reproducible experimental conditions. |
| Researcher Affiliation | Collaboration | 1 Department of Computer Science, ETH Zurich 2 Max Planck Institute for Intelligent Systems, T ubingen 3 Google Research, Brain Team |
| Pseudocode | No | The paper describes models and training procedures but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks, nor does it present structured steps formatted like code. |
| Open Source Code | No | The paper does not contain an explicit statement about releasing source code for the methodology or a link to a code repository. |
| Open Datasets | Yes | We consider four commonly used disentanglement data sets where one has explicit access to the ground-truth generative model and the factors of variation: d Sprites (Higgins et al., 2017a), Cars3D (Reed et al., 2015), Small NORB (Le Cun et al., 2004) and Shapes3D (Kim & Mnih, 2018). |
| Dataset Splits | Yes | To include supervision during training we split the labeled examples in a 90%/10% train/validation split. |
| Hardware Specification | Yes | Reproducing these experiment requires approximately 8.57 GPU years (NVIDIA P100). |
| Software Dependencies | No | The paper mentions using 'the disentanglement lib' but does not specify its version. It also lists 'Adam' as an optimizer with specific hyperparameters, but no version number for Adam or any other software libraries or programming languages are provided. |
| Experiment Setup | Yes | For a detailed description of hyperparameters, architecture, and model training we refer to Appendix B. Table 3: Hyperparameters explored for the different disentanglement methods. Table 4: Other fixed hyperparameters (e.g., Batch size 64, Latent space dimension 10, Optimizer Adam, Adam: learning rate 0.0001, Training steps 300000). |