Releases: douglasrizzo/hcanet
AAAI-RLG (refined)
This release contains the code for work presented as part of my thesis. It is also a less janky version of the code used in the experiments presented in the paper below. With this release, it is still possible to run experiments akin to the ones in the v8 tag, but faster, as long as the encoding and action selection modules aren't shared among all agents. I also believe temporal relation regularization is not implemented anymore, which is present in v8, but did not have good results.
Meneghetti, D. D. R., & Bianchi, R. A. da C. (2021). Specializing Inter-Agent Communication in Heterogeneous Multi-Agent Reinforcement Learning using Agent Class Information. AAAI-21 Workshop on Reinforcement Learning in Games, 12. https://arxiv.org/abs/2012.07617
@inproceedings{Meneghetti2021a,
title = {Specializing {{Inter}}-{{Agent Communication}} in {{Heterogeneous Multi}}-{{Agent Reinforcement Learning}} Using {{Agent Class Information}}},
booktitle = {{{AAAI}}-21 {{Workshop}} on {{Reinforcement Learning}} in {{Games}}},
author = {Meneghetti, Douglas De Rizzo and Bianchi, Reinaldo Augusto da Costa},
year = {2021},
month = feb,
pages = {12},
address = {{Virtual Conference}},
copyright = {All rights reserved},
url = {https://arxiv.org/abs/2012.07617},
language = {en-US}
}ENIAC paper
This release contains the code for the work presented in the following paper.
Meneghetti, D. D. R., & Bianchi, R. A. da C. (2020). Towards Heterogeneous Multi-Agent Reinforcement Learning with Graph Neural Networks. Anais Do Encontro Nacional de Inteligência Artificial e Computacional (ENIAC), 579–590. https://doi.org/10.5753/eniac.2020.12161
@inproceedings{Meneghetti2020b,
title = {Towards {{Heterogeneous Multi}}-{{Agent Reinforcement Learning}} with {{Graph Neural Networks}}},
booktitle = {Anais Do {{Encontro Nacional}} de {{Intelig\^encia Artificial}} e {{Computacional}} ({{ENIAC}})},
author = {Meneghetti, Douglas De Rizzo and Bianchi, Reinaldo Augusto da Costa},
year = {2020},
month = oct,
pages = {579--590},
publisher = {{SBC}},
address = {{Porto Alegre, RS, Brasil}},
doi = {10.5753/eniac.2020.12161},
abstract = {This work proposes a neural network architecture that learns policies for multiple agent classes in a heterogeneous multi-agent reinforcement setting. The proposed network uses directed labeled graph representations for states, encodes feature vectors of different sizes for different entity classes, uses relational graph convolution layers to model different communication channels between entity types and learns distinct policies for different agent classes, sharing parameters wherever possible. Results have shown that specializing the communication channels between entity classes is a promising step to achieve higher performance in environments composed of heterogeneous entities.},
language = {en}
}