Symmetry-Based Disentangled Representation Learning requires Interaction with Environments
Authors: Hugo Caselles-Dupré, Michael Garcia Ortiz, David Filliat
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experiments can be reproduced in Colab 1 and the code is available on Git Hub 2. We build on their work and make observations, theoretical and empirical, that lead us to argue that SBDRL requires interaction with environments. We propose empirical implementations that are able to successfully approximate these analytically defined representations. We empirically demonstrate the efficiency of using SB-disentangled for a downstream task (learning an inverse model). |
| Researcher Affiliation | Collaboration | Hugo Caselles-Dupré1,2, Michael Garcia-Ortiz2, David Filliat1 1Flowers Laboratory (ENSTA Paris & INRIA), 2AI Lab (Softbank Robotics Europe) caselles@ensta.fr, mgarciaortiz@softbankrobotics.com, david.filliat@ensta.fr |
| Pseudocode | Yes | The training procedure is presented in Algorithm 1 in Appendix C.2 |
| Open Source Code | Yes | Our experiments can be reproduced in Colab 1 and the code is available on Git Hub 2. 2https://github.com/Caselles/Neur IPS19-SBDRL 4https://github.com/Caselles/Neur IPS19-SBDRL |
| Open Datasets | Yes | We implement this simple environment using Flatland (Caselles-Dupré et al., 2018). We collect 10k transitions (ot, at, ot+1). |
| Dataset Splits | Yes | We then report the 10-fold cross-validation mean accuracy as a function of the maximum depth parameter of random forest, which controls the capacity of the classifier. |
| Hardware Specification | No | The paper mentions that experiments can be reproduced in 'Colab' but does not provide any specific hardware details such as GPU/CPU models or memory. |
| Software Dependencies | No | The paper mentions 'Scikit-learn (Pedregosa et al., 2011)' and 'Pytorch (Paszke et al., 2017)' but does not provide specific version numbers for these software dependencies. |
| Experiment Setup | No | All architecture and hyperparameters details are specified in Appendix B. This statement indicates that the specific experimental setup details are not provided in the main text. |