From de84c41f2a9c9e5adf5d193fe97974c7432b9302 Mon Sep 17 00:00:00 2001 From: Nemo Date: Thu, 26 Nov 2020 14:28:09 +0530 Subject: [PATCH] New Hanabi papers --- README.md | 4 + boardgame-research.bib | 71 +++++++++ boardgame-research.rdf | 329 +++++++++++++++++++++++++++++++++++++++++ 3 files changed, 404 insertions(+) diff --git a/README.md b/README.md index 5cc50fe..bf1db86 100644 --- a/README.md +++ b/README.md @@ -99,6 +99,10 @@ There is a [simulator](https://dominionsimulator.wordpress.com/f-a-q/) and the c - [A strategy simulator for the well-known cooperative card game Hanabi](https://github.com/rjtobin/HanSim) - [A framework for writing bots that play Hanabi](https://github.com/Quuxplusone/Hanabi) - [Evaluating the Rainbow DQN Agent in Hanabi with Unseen Partners](https://arxiv.org/abs/2004.13291) +- [Operationalizing Intentionality to Play Hanabi with Human Players](https://doi.org/10.1109/TG.2020.3009359) +- [Behavioral Evaluation of Hanabi Rainbow DQN Agents and Rule-Based Agents](https://ojs.aaai.org//index.php/AIIDE/article/view/7404) [[pdf](https://ojs.aaai.org/index.php/AIIDE/article/view/7404/7333)] +- [Playing mini-Hanabi card game with Q-learning](http://id.nii.ac.jp/1001/00205046/) +- [Generating and Adapting to Diverse Ad-Hoc Cooperation Agents in Hanabi](https://arxiv.org/abs/2004.13710) # Hive diff --git a/boardgame-research.bib b/boardgame-research.bib index 7d9c60d..cfc8a66 100644 --- a/boardgame-research.bib +++ b/boardgame-research.bib @@ -1012,3 +1012,74 @@ url = {https://www.rtealwitter.com/slides/2020-JMM.pdf}, author = {Witter, R. Teal and Lyford, Alex} } + +@article{gendre_playing_2020, + title = {Playing {Catan} with {Cross}-dimensional {Neural} {Network}}, + url = {http://arxiv.org/abs/2008.07079}, + abstract = {Catan is a strategic board game having interesting properties, including multi-player, imperfect information, stochastic, complex state space structure (hexagonal board where each vertex, edge and face has its own features, cards for each player, etc), and a large action space (including negotiation). Therefore, it is challenging to build AI agents by Reinforcement Learning (RL for short), without domain knowledge nor heuristics. In this paper, we introduce cross-dimensional neural networks to handle a mixture of information sources and a wide variety of outputs, and empirically demonstrate that the network dramatically improves RL in Catan. We also show that, for the first time, a RL agent can outperform jsettler, the best heuristic agent available.}, + urldate = {2020-10-12}, + journal = {arXiv:2008.07079 [cs, stat]}, + author = {Gendre, Quentin and Kaneko, Tomoyuki}, + month = aug, + year = {2020}, + note = {arXiv: 2008.07079}, + keywords = {Computer Science - Artificial Intelligence, Computer Science - Machine Learning, Statistics - Machine Learning}, + annote = {Comment: 12 pages, 5 tables and 10 figures; submitted to the ICONIP 2020}, + file = {arXiv Fulltext PDF:/home/nemo/Zotero/storage/AU6NYDIV/Gendre and Kaneko - 2020 - Playing Catan with Cross-dimensional Neural Networ.pdf:application/pdf;arXiv.org Snapshot:/home/nemo/Zotero/storage/NKRW6UKC/2008.html:text/html} +} + +@inproceedings{theodoridis_monte_2020, + address = {Athens Greece}, + title = {Monte {Carlo} {Tree} {Search} for the {Game} of {Diplomacy}}, + isbn = {978-1-4503-8878-8}, + url = {https://dl.acm.org/doi/10.1145/3411408.3411413}, + doi = {10.1145/3411408.3411413}, + language = {en}, + urldate = {2020-10-12}, + booktitle = {11th {Hellenic} {Conference} on {Artificial} {Intelligence}}, + publisher = {ACM}, + author = {Theodoridis, Alexios and Chalkiadakis, Georgios}, + month = sep, + year = {2020}, + pages = {16--25} +} + +@article{eger_operationalizing_2020, + title = {Operationalizing {Intentionality} to {Play} {Hanabi} with {Human} {Players}}, + issn = {2475-1502, 2475-1510}, + url = {https://ieeexplore.ieee.org/document/9140404/}, + doi = {10.1109/TG.2020.3009359}, + urldate = {2020-11-26}, + journal = {IEEE Transactions on Games}, + author = {Eger, Markus and Martens, Chris and Sauma Chacon, Pablo and Alfaro Cordoba, Marcela and Hidalgo Cespedes, Jeisson}, + year = {2020}, + pages = {1--1}, + file = {Full Text:/home/nemo/Zotero/storage/V2M3QSJG/Eger et al. - 2020 - Operationalizing Intentionality to Play Hanabi wit.pdf:application/pdf} +} + +@article{canaan_behavioral_2020, + title = {Behavioral {Evaluation} of {Hanabi} {Rainbow} {DQN} {Agents} and {Rule}-{Based} {Agents}}, + volume = {16}, + url = {https://ojs.aaai.org/index.php/AIIDE/article/view/7404}, + abstract = {\<p class=\"abstract\"\>Hanabi is a multiplayer cooperative card game, where only your partners know your cards. All players succeed or fail together. This makes the game an excellent testbed for studying collaboration. Recently, it has been shown that deep neural networks can be trained through self-play to play the game very well. However, such agents generally do not play well with others. In this paper, we investigate the consequences of training Rainbow DQN agents with human-inspired rule-based agents. We analyze with which agents Rainbow agents learn to play well, and how well playing skill transfers to agents they were not trained with. We also analyze patterns of communication between agents to elucidate how collaboration happens. A key finding is that while most agents only learn to play well with partners seen during training, one particular agent leads the Rainbow algorithm towards a much more general policy. The metrics and hypotheses advanced in this paper can be used for further study of collaborative agents.\</p\>}, + number = {1}, + urldate = {2020-11-26}, + journal = {Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment}, + author = {Canaan, Rodrigo and Gao, Xianbo and Chung, Youjin and Togelius, Julian and Nealen, Andy and Menzel, Stefan}, + month = oct, + year = {2020}, + note = {Section: Full Oral Papers}, + pages = {31--37} +} + +@inproceedings{_playing_2020, + title = {Playing mini-{Hanabi} card game with {Q}-learning}, + volume = {2020}, + url = {http://id.nii.ac.jp/1001/00205046/}, + booktitle = {第82回全国大会講演論文集}, + author = {ひい, とう and 市来, 正裕 and 中里, 研一}, + month = feb, + year = {2020}, + note = {Issue: 1}, + pages = {41--42} +} diff --git a/boardgame-research.rdf b/boardgame-research.rdf index 66b202b..77ec343 100644 --- a/boardgame-research.rdf +++ b/boardgame-research.rdf @@ -4335,6 +4335,330 @@ DOI: 10.1007/978-3-319-71649-7_5 + + journalArticle + + + arXiv:2008.07079 [cs, stat] + + + + + + + Gendre + Quentin + + + + + Kaneko + Tomoyuki + + + + + + + + + + Computer Science - Artificial Intelligence + + + + + Computer Science - Machine Learning + + + + + Statistics - Machine Learning + + + Playing Catan with Cross-dimensional Neural Network + Catan is a strategic board game having interesting properties, including multi-player, imperfect information, stochastic, complex state space structure (hexagonal board where each vertex, edge and face has its own features, cards for each player, etc), and a large action space (including negotiation). Therefore, it is challenging to build AI agents by Reinforcement Learning (RL for short), without domain knowledge nor heuristics. In this paper, we introduce cross-dimensional neural networks to handle a mixture of information sources and a wide variety of outputs, and empirically demonstrate that the network dramatically improves RL in Catan. We also show that, for the first time, a RL agent can outperform jsettler, the best heuristic agent available. + 2020-08-17 + arXiv.org + + + http://arxiv.org/abs/2008.07079 + + + 2020-10-12 04:19:57 + arXiv: 2008.07079 + + + Comment: 12 pages, 5 tables and 10 figures; submitted to the ICONIP 2020 + + + attachment + arXiv Fulltext PDF + + + https://arxiv.org/pdf/2008.07079.pdf + + + 2020-10-12 04:20:04 + 1 + application/pdf + + + attachment + arXiv.org Snapshot + + + https://arxiv.org/abs/2008.07079 + + + 2020-10-12 04:20:10 + 1 + text/html + + + conferencePaper + + + ISBN 978-1-4503-8878-8 + 11th Hellenic Conference on Artificial Intelligence + DOI 10.1145/3411408.3411413 + + + + + + + Athens Greece + + + ACM + + + + + + + Theodoridis + Alexios + + + + + Chalkiadakis + Georgios + + + + + Monte Carlo Tree Search for the Game of Diplomacy + 2020-09-02 + en + DOI.org (Crossref) + + + https://dl.acm.org/doi/10.1145/3411408.3411413 + + + 2020-10-12 04:20:38 + 16-25 + + + SETN 2020: 11th Hellenic Conference on Artificial Intelligence + + + + + journalArticle + + + + + + Eger + Markus + + + + + Martens + Chris + + + + + Sauma Chacon + Pablo + + + + + Alfaro Cordoba + Marcela + + + + + Hidalgo Cespedes + Jeisson + + + + + + Operationalizing Intentionality to Play Hanabi with Human Players + 2020 + DOI.org (Crossref) + + + https://ieeexplore.ieee.org/document/9140404/ + + + 2020-11-26 08:48:44 + 1-1 + + + IEEE Transactions on Games + DOI 10.1109/TG.2020.3009359 + IEEE Trans. Games + ISSN 2475-1502, 2475-1510 + + + attachment + Full Text + + + https://sci-hub.se/downloads/2020-08-17/f1/eger2020.pdf?rand=5fbf6bef76c6b#view=FitH + + + 2020-11-26 08:48:52 + 1 + application/pdf + + + journalArticle + + + 16 + Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment + 1 + AIIDE + + + + + + + Canaan + Rodrigo + + + + + Gao + Xianbo + + + + + Chung + Youjin + + + + + Togelius + Julian + + + + + Nealen + Andy + + + + + Menzel + Stefan + + + + + + Behavioral Evaluation of Hanabi Rainbow DQN Agents and Rule-Based Agents + &lt;p class=&quot;abstract&quot;&gt;Hanabi is a multiplayer cooperative card game, where only your partners know your cards. All players succeed or fail together. This makes the game an excellent testbed for studying collaboration. Recently, it has been shown that deep neural networks can be trained through self-play to play the game very well. However, such agents generally do not play well with others. In this paper, we investigate the consequences of training Rainbow DQN agents with human-inspired rule-based agents. We analyze with which agents Rainbow agents learn to play well, and how well playing skill transfers to agents they were not trained with. We also analyze patterns of communication between agents to elucidate how collaboration happens. A key finding is that while most agents only learn to play well with partners seen during training, one particular agent leads the Rainbow algorithm towards a much more general policy. The metrics and hypotheses advanced in this paper can be used for further study of collaborative agents.&lt;/p&gt; + October 1, 2020 + + + https://ojs.aaai.org/index.php/AIIDE/article/view/7404 + + + 2020-11-26 + Section: Full Oral Papers + 31-37 + + + attachment + View PDF + + + https://ojs.aaai.org/index.php/AIIDE/article/view/7404/7333 + + + 2020-11-26 08:52:38 + 3 + + + conferencePaper + + + 2020第82回全国大会講演論文集 + + + + + + + ひい + とう + + + + + 市来 + 正裕 + + + + + 中里 + 研一 + + + + + + Playing mini-Hanabi card game with Q-learning + February 2020 + + + http://id.nii.ac.jp/1001/00205046/ + + + Issue: 1 + 41–42 + + + attachment + View PDF + + + https://ipsj.ixsq.nii.ac.jp/ej/?action=repository_uri&item_id=205142&file_id=1&file_no=1 + + + 2020-11-26 08:54:47 + 3 + Accessibility @@ -4345,6 +4669,7 @@ DOI: 10.1007/978-3-319-71649-7_5 + Dixit @@ -4372,6 +4697,9 @@ DOI: 10.1007/978-3-319-71649-7_5 + + + Hive @@ -4487,6 +4815,7 @@ DOI: 10.1007/978-3-319-71649-7_5 + Shobu