Geometric Inductive Biases for Identifiable Unsupervised Learning of Disentangled Representations
Authors: Ziqi Pan, Li Niu, Liqing Zhang
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The experimental results show the usefulness of the geometric inductive biases in unsupervised disentangling and the effectiveness of our GDRAE in capturing the geometric inductive biases. |
| Researcher Affiliation | Academia | Mo E Key Lab of Artificial Intelligence, Department of Computer Science and Engineering Shanghai Jiao Tong University, Shanghai, China {panziqi ai, ustcnewly}@sjtu.edu.cn, zhang-lq@cs.sjtu.edu.cn |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper provides a link to the 3DShapes dataset ('https: //github.com/deepmind/3d-shapes/'), which is a third-party resource, but does not state that the authors' own source code for their methodology is released or available. |
| Open Datasets | Yes | For the 3DShapes (Burgess and Kim 2019) dataset, samples varying only in floor hue, wall hue and object hue can be depicted by an α-structure manifold, since spatial areas affected by these factors are also disjoint (see Fig. 1(b)). We consider images datasets D = x RH W C . For certain generative factors of certain datasets, different factors only affect disjoint subspaces of RH W C, so in such a case Eq. (4) is permitted when different latent dimensions zj align with different factors. For example, given a human face image from the Celeb A (Liu et al. 2015) dataset, the factors of smile degree and hair color may affect disjoint subspaces of the image, since the spatial areas of mouth and hair are disjoint. As shown by (Shao, Kumar, and Thomas Fletcher 2018), for images a, b and c from real data manifold such as Celeb A (Liu et al. 2015) and SVHN (Netzer et al. 2011), parallel transport finds an image d that is related to c in the same semantic manner as a is related to b (Shao, Kumar, and Thomas Fletcher 2018). |
| Dataset Splits | No | The paper mentions using datasets but does not provide specific details about training, validation, or test dataset splits (e.g., percentages, sample counts, or predefined split references). |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used to run the experiments. |
| Software Dependencies | No | The paper does not list specific software dependencies with version numbers (e.g., 'PyTorch 1.9' or 'Python 3.8'). |
| Experiment Setup | No | The paper mentions 'For both models, we set K = 2' and 'tuned hyper-parameters for training β-VAE', indicating some setup. It also states 'The model is constituted as in Eq. (4) (see supplementary for the detailed process of model construction)'. However, it does not provide specific hyperparameter values (e.g., learning rate, batch size) or other concrete training configurations within the main text. |