Interesting Object, Curious Agent: Learning Task-Agnostic Exploration

Authors: Simone Parisi, Victoria Dean, Deepak Pathak, Abhinav Gupta

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments are designed to highlight the benefits of disentangling the environment-centric nature of exploration from agent-centric behavior by learning a separate exploration policy and then transferring it to new environments. Environments. The experiments are divided into two main sections. The first is about Mini Grid [10] (Section 4.1), a set of procedurally-generated environments where the agent can interact with many objects. The second is about Habitat [46] (Section 4.2), a navigation simulator showcasing the generality of our Mini Grid experiments to a visually realistic domain.
Researcher Affiliation Collaboration Simone Parisi1 Victoria Dean2 Deepak Pathak2 Abhinav Gupta1 1Facebook AI Research 2Carnegie Mellon University
Pseudocode No The paper describes processes and frameworks through text and diagrams (e.g., Figures 2 and 3) but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes The source code is available at https://github.com/sparisi/cbet/.
Open Datasets Yes The experiments are divided into two main sections. The first is about Mini Grid [10]... The second is about Habitat [46]... To demonstrate that C-BET s efficacy extends to realistic settings with visual inputs, we perform experiments on Habitat [46] with Replica scenes [52].
Dataset Splits No The paper mentions 'training-testing environment pairs' and states 'In Appendix A.2 we specify all the training details (e.g., data splits, hyperparameters, how they were chosen).' However, the provided text does not include Appendix A.2, so specific split percentages or sample counts for training/validation/test splits are not available within the main body.
Hardware Specification No The paper states 'We specify all the compute details in Appendix A.3.' in its author checklist. However, Appendix A.3 is not included in the provided text, and thus, no specific hardware details (e.g., GPU/CPU models, memory) are available in the main body of the paper.
Software Dependencies No The paper mentions software components such as 'IMPALA [16]' and '#Exploration [54]' but does not provide specific version numbers for these or other software dependencies, which are required for a reproducible description.
Experiment Setup No The paper mentions 'In Appendix A.2 we specify all the training details (e.g., data splits, hyperparameters, how they were chosen).' However, the provided text does not include Appendix A.2, and therefore, specific hyperparameter values or detailed training configurations are not available in the main body of the paper.