Gated Path Planning Networks
Authors: Lisa Lee, Emilio Parisotto, Devendra Singh Chaplot, Eric Xing, Ruslan Salakhutdinov
ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this work, we reframe VINs as recurrent-convolutional networks which demonstrates that VINs couple recurrent convolutions with an unconventional max-pooling activation. From this perspective, we argue that standard gated recurrent update equations could potentially alleviate the optimization issues plaguing VIN. The resulting architecture, which we call the Gated Path Planning Network, is shown to empirically outperform VIN on a variety of metrics such as learning speed, hyperparameter sensitivity, iteration count, and even generalization. |
| Researcher Affiliation | Academia | 1Carnegie Mellon University, Machine Learning Department. |
| Pseudocode | No | The paper provides mathematical equations for model updates but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any explicit statements about releasing source code for the methodology described, nor does it include a link to a code repository. |
| Open Datasets | No | The paper describes how the maze environments and corresponding data were generated ('The 2D maze environment is created with a maze generation process that uses Depth-First Search with the Recursive Backtracker algorithm (Maze Generation Algorithms, 2018)', 'For each Doom maze, we take RGB screenshots showing the first-person view of the environment'), but it does not provide concrete access information (link, DOI, citation to a specific dataset) for the publicly available datasets used or generated for the experiments. |
| Dataset Splits | Yes | Unless otherwise noted, the results were obtained by doing a hyperparameter sweep of (K, F) over K 2 {5, 10, 15, 20, 30} and F 2 {3, 5, 7, 9, 11}, and using a 25k/5k/5k train-val-test split. |
| Hardware Specification | No | The paper thanks Nvidia for providing GPU support ('The authors would also like to thank Nvidia for providing GPU support.') but does not specify any exact GPU models, CPU models, memory details, or other specific hardware configurations used for running the experiments. |
| Software Dependencies | No | The paper mentions using 'the Doom Game Engine and the Vi ZDoom API (Kempka et al., 2016)' but does not provide specific version numbers for these or any other software libraries or frameworks (e.g., PyTorch, TensorFlow, etc.) used in the experiments. |
| Experiment Setup | Yes | Unless otherwise noted, the results were obtained by doing a hyperparameter sweep of (K, F) over K 2 {5, 10, 15, 20, 30} and F 2 {3, 5, 7, 9, 11}, and using a 25k/5k/5k train-val-test split. Other experimental details are deferred to the Appendix. ... The results were obtained using K = 30, the best setting of F for each transition kernel, a smaller dataset size 10k (due to memory and time constraints), a smaller learning rate 5e-4, and 100 training epochs. |