Simulating Action Dynamics with Neural Process Networks
Authors: Antoine Bosselut, Omer Levy, Ari Holtzman, Corin Ennis, Dieter Fox, Yejin Choi
ICLR 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical results demonstrate that our proposed model can reason about the unstated causal effects of actions, allowing it to provide more accurate contextual information for understanding and generating procedural text, all while offering more interpretable internal representations than existing alternatives. |
| Researcher Affiliation | Academia | Paul G. Allen School of Computer Science & Engineering University of Washington {antoineb,omerlevy,ahai,fox,yejin}@cs.washington.edu School of Science, Technology, Engineering & Mathematics University of Washington Bothell {corin123}@uw.edu |
| Pseudocode | No | The paper describes the model's architecture and processes using equations and text, but it does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper states, 'Finally, we include a new dataset with fine-grained annotations on state changes, to be shared publicly, to encourage future research in this direction.' It does not explicitly mention the release of source code for the methodology. |
| Open Datasets | Yes | For learning and evaluation, we use a subset of the Now You re Cooking dataset (Kiddon et al., 2016). |
| Dataset Splits | Yes | We chose 65816 recipes for training, 175 recipes for development, and 700 recipes for testing. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU or CPU models used for running the experiments. It only mentions general settings like 'The hidden size of the instruction encoder is 100...' |
| Software Dependencies | No | The paper mentions using components like Gated Recurrent Unit (Cho et al., 2014) and the Adam optimizer (Kingma & Ba, 2014), but it does not specify software dependencies with version numbers such as PyTorch, TensorFlow, or specific library versions. |
| Experiment Setup | Yes | The hidden size of the instruction encoder is 100, the embedding sizes of action functions and entities are 30. We use dropout with a rate of 0.3 before any non-recurrent fully connected layers. We use the Adam optimizer with a learning rate of .001 and decay by a factor of 0.1 if we see no improvement on validation loss over three epochs. The batch size is 64. |