Don’t Do What Doesn’t Matter: Intrinsic Motivation with Action Usefulness

Authors: Mathieu Seurin, Florian Strub, Philippe Preux, Olivier Pietquin

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate Do Wha M on the procedurally-generated environment Mini Grid, against state-of-the-art methods. Experiments consistently show that Do Wha M greatly reduces sample complexity, installing the new state-of-the-art in Mini Grid.
Researcher Affiliation Collaboration 1Univ. Lille, CNRS, Inria, Centrale Lille, UMR 9189 CRISt AL, F-59000 Lille, France 2Deep Mind, Paris, France 3Google Research, Brain Team, Paris, France
Pseudocode No The paper describes the method in prose and mathematical formulas, but does not include a clearly labeled pseudocode or algorithm block.
Open Source Code Yes Open-Source code available at: https://github.com/Mathieu-Seurin/impact-driven-exploration
Open Datasets Yes We evaluate Do Wha M in the procedurally-generated environments Mini Grid [Chevalier-Boisvert et al., 2018].
Dataset Splits No The paper states 'We follow the training protocol defined in [Raileanu and Rockt aschel, 2019; Campero et al., 2020]' but does not explicitly provide training/validation/test dataset splits for the procedurally-generated Mini Grid environment.
Hardware Specification No The paper states 'Experiments presented in this paper were carried out using the Grid 5000 testbed,' but does not provide specific hardware details such as GPU or CPU models, or memory specifications.
Software Dependencies No The paper mentions using 'IMPALA [Espeholt et al., 2018] Torch Beast implementation [Raileanu and Rockt aschel, 2019]' but does not provide specific version numbers for the software dependencies like PyTorch or Torch Beast.
Experiment Setup Yes We use 3 convolution layers with a kernel size of 3, followed by 2 fully-connected layers of size 1024, and an LSTM of hidden size 1024. Finally, we use two separate fully-connected layers of size 1024 for the actor s and critic s head.