Introducing Symmetries to Black Box Meta Reinforcement Learning

Authors: Louis Kirsch, Sebastian Flennerhag, Hado van Hasselt, Abram Friesen, Junhyuk Oh, Yutian Chen7202-7210

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

Reproducibility Variable Result LLM Response
Research Type Experimental We experimentally validate Sym LA on bandits, classic control, and grid worlds, comparing generalisation capabilities to Meta RNNs. Sym LA improves generalisation when varying action dimensions, permuting observations and actions, and significantly changing tasks and environments.
Researcher Affiliation Collaboration Louis Kirsch1 2, Sebastian Flennerhag1, Hado van Hasselt1, Abram Friesen1, Junhyuk Oh1, Yutian Chen1 1 DeepMind 2 The Swiss AI Lab IDSIA, USI, SUPSI
Pseudocode Yes For pseudo code, see Algorithm 1 in the appendix.
Open Source Code No The paper does not contain any explicit statement about releasing source code for the described methodology, nor does it provide a link to a code repository.
Open Datasets Yes We first compare Sym LA and the Meta RNN on the two-armed (dependent) bandit experiments from Wang et al. (2016)
Dataset Splits No The paper discusses 'meta-training' and 'meta-testing' environments and 'meta-training distribution' but does not specify numerical train/validation/test splits (e.g., percentages or sample counts) for the datasets used.
Hardware Specification No The paper does not provide specific hardware details such as GPU or CPU models, or cloud computing specifications used for running the experiments.
Software Dependencies No The paper does not list specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions) that would be required for replication.
Experiment Setup Yes Hyper-parameters are in Appendix B.