Enriching Disentanglement: From Logical Definitions to Quantitative Metrics
Authors: Yivan Zhang, Masashi Sugiyama
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we empirically demonstrate the effectiveness of the proposed metrics by isolating different aspects of disentangled representations. |
| Researcher Affiliation | Academia | Yivan Zhang The University of Tokyo, RIKEN AIP Tokyo, Japan yivanzhang@ms.k.u-tokyo.ac.jp Masashi Sugiyama RIKEN AIP, The University of Tokyo Tokyo, Japan sugi@k.u-tokyo.ac.jp |
| Pseudocode | No | The paper includes Python code snippets in Appendix D.6 (e.g., 'def q_product(y: np.ndarray, z: np.ndarray, aggregate, deviation):'), but these are embedded in text for illustration rather than presented as formally labeled 'Pseudocode' or 'Algorithm' blocks or figures. |
| Open Source Code | No | The paper provides code snippets for implementing metrics in Appendix D.6 and mentions using a 'public PyTorch implementation' for existing models, but it does not include an unambiguous statement of releasing its own source code for the described methodology or a direct link to a dedicated repository for its novel contributions. |
| Open Datasets | Yes | We also report the results of several widely used unsupervised disentangled representation learning methods (VAE [Kingma and Welling, 2014], β-VAE [Higgins et al., 2017], Factor VAE [Kim and Mnih, 2018], and β-TCVAE [Chen et al., 2018]) evaluated on four image datasets (3D Cars [Reed et al., 2015], d Sprites [Matthey et al., 2017], 3D Shapes [Burgess and Kim, 2018], and MPI3D [Gondal et al., 2019]). |
| Dataset Splits | No | The paper states, 'We used a public Py Torch implementation ... and used the same encoder/decoder architecture with the default hyperparameters described in Locatello et al. [2019b] for all methods for a fair comparison.' While it refers to external hyperparameters, it does not explicitly provide the training/validation/test dataset splits within the paper itself. |
| Hardware Specification | Yes | The experiments were conducted on a NVIDIA Tesla V100 GPU. |
| Software Dependencies | No | The paper mentions software like Num Py [Harris et al., 2020], Py Torch [Paszke et al., 2019], and scikit-learn [Pedregosa et al., 2011], but it does not provide specific version numbers for these key software components required for reproducibility. |
| Experiment Setup | Yes | We used a public Py Torch implementation [Paszke et al., 2019] of these methods and used the same encoder/decoder architecture with the default hyperparameters described in Locatello et al. [2019b] for all methods for a fair comparison. |