Homomorphism AutoEncoder -- Learning Group Structured Representations from Observed Transitions
Authors: Hamza Keurti, Hsiao-Ru Pan, Michel Besserve, Benjamin F Grewe, Bernhard Schölkopf
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We motivate our method theoretically, and show empirically1 that it can learn a group representation of the actions, thereby capturing the structure of the set of transformations applied to the environment. We further show that this allows agents to predict the effect of sequences of future actions with improved accuracy. |
| Researcher Affiliation | Academia | 1Max Planck Institute for Intelligent Systems, T ubingen, Germany 2Institute of Neuroinformatics, ETH Z urich, Switzerland 3Max Planck ETH Center for Learning Systems. |
| Pseudocode | No | The paper describes the architecture and training process in text and diagrams, but does not provide structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | 1Code can be found at https://github.com/ hamzakeurti/homomorphismvae |
| Open Datasets | Yes | We consider a subset of the d Sprites dataset (Matthey et al., 2017) and Matthey, L., Higgins, I., Hassabis, D., and Lerchner, A. dsprites: Disentanglement testing sprites dataset. https://github.com/deepmind/dsprites-dataset/, 2017. For the 3D rotation experiment, we used the Stanford bunny (Turk & Levoy, 1994) |
| Dataset Splits | No | We evaluate the methods in an offline setting, where we train each method on a given set of 2-step trajectories and test their generalization ability on a hold-out set of 128-step trajectories. This only mentions training and test sets, not a validation set or its split details. |
| Hardware Specification | Yes | The experiments were performed on an NVIDIA Ge Force RTX 3090 and A100 GPUs |
| Software Dependencies | No | To implement our architecture we used the deep learning framework Py Torch. (Paszke et al., 2019). This mentions PyTorch but not its specific version number, nor other software with versions. |
| Experiment Setup | Yes | Architecture and hyperparameters for training are specified in the appendix C. and includes tables Table 2. Network architecture. and Table 3. Training hyperparameters. listing specific values for various parameters (e.g., Learning rate 0.001, Batch size 500, Epochs 101, Latent space 4, γ 400). |