Interaction-Based Disentanglement of Entities for Object-Centric World Models
Authors: Akihiro Nakano, Masahiro Suzuki, Yutaka Matsuo
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical evaluation shows that the proposed model factorizes global features, unaffected by interactions from relational features that are necessary to predict the outcome of interactions. We also show that STEDIE achieves better performance in planning tasks and understanding causal relationships. In both tasks, our model not only achieves better performance in terms of reconstruction ability but also utilizes the disentangled representations to solve the tasks in a structured manner. |
| Researcher Affiliation | Academia | Akihiro Nakano, Masahiro Suzuki, and Yutaka Matsuo Graduate School of Engineering The University of Tokyo 7-3-1 Hongo, Bunkyo-ku, Tokyo-to, Japan {nakano.akihiro,masa,matsuo}@weblab.t.u-tokyo.ac.jp |
| Pseudocode | No | Figure 1 (a) shows the graphical model of our proposed model, STEDIE. |
| Open Source Code | No | The paper does not contain any explicit statement about providing open-source code for the described methodology, nor does it provide a link to a code repository. |
| Open Datasets | Yes | We evaluate STEDIE using two datasets each with different characteristics: (1) 3D block moving dataset (Blocks) from Janner et al. (2019) and Veerapaneni et al. (2020), and (2) 2D weighted-block pushing dataset (Weights) from Ke et al. (2021). |
| Dataset Splits | No | To train OP3, we conducted a grid search on the hyperparameters with batch size of [16, 32, 80], learning rate of [0.0001, 0.0003], and KL coefficient of [0.001, 0.01]. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as GPU models, CPU types, or memory used for running the experiments. |
| Software Dependencies | No | The paper mentions software components like 'Adam optimizer', 'MLP', and 'CNN' but does not provide specific version numbers for any programming languages, libraries, or solvers. |
| Experiment Setup | Yes | We trained the model with a fixed learning rate of 0.0005 for 400 epochs. The coefficients of KL divergence terms in the loss function were set to 0.01 for relational feature and 0.1 for global feature. Iterative inference was conducted with 4 refinement steps. The number of slots were set to 4. |