Constrained GPI for Zero-Shot Transfer in Reinforcement Learning

Authors: Jaekyeom Kim, Seohong Park, Gunhee Kim

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through experiments in the Scavenger and Reacher environment with state observations as well as the Deep Mind Lab environment with visual observations, we show that the proposed constrained GPI significantly outperforms the prior GPI s transfer performance.
Researcher Affiliation Academia Jaekyeom Kim Seoul National University jaekyeom@snu.ac.kr Seohong Park University of California, Berkeley seohong@berkeley.edu Gunhee Kim Seoul National University gunhee@snu.ac.kr
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes Our code and additional information are available at https://jaekyeom.github.io/projects/cgpi/.
Open Datasets Yes We start our experiments in the Scavenger environment [8, 9],... and the Deep Mind Lab environment [7, 10, 12].
Dataset Splits No The paper mentions training and testing but does not provide specific details on dataset splits for training, validation, and testing (e.g., percentages or sample counts for each split).
Hardware Specification No The paper states: "Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [Yes] We will include in the supplementary material." This information is not present in the main paper.
Software Dependencies No The paper mentions Pytorch [27] and Adam [21] but does not specify their version numbers or other software dependencies with specific versions.
Experiment Setup No The paper states: "Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [Yes] We will include in the supplementary material." This information is not present in the main paper.