Behavioral Differences is the Key of Ad-hoc Team Cooperation in Multiplayer Games Hanabi

Post hoc Post-hoc analysis
DOI: 10.48550/arxiv.2303.06775 Publication Date: 2023-01-01
ABSTRACT
Ad-hoc team cooperation is the problem of cooperating with other players that have not been seen in learning process. Recently, this has considered context Hanabi, which requires without explicit communication players. While self-play strategies on reinforcement (RL) process shown success, there failing to cooperate unseen agents after initial completed. In paper, we categorize results ad-hoc into Failure, Success, and Synergy analyze associated failures. First, confirm via RL converge one strategy each, but necessarily same these can deploy different even though they utilize hyperparameters. Second, larger behavioral difference, more pronounced failure cooperation, as demonstrated using hierarchical clustering Pearson correlation. We such are grouped distinctly groups through clustering, correlation between differences performance -0.978. Our improve understanding key factors form successful multi-player games.
SUPPLEMENTAL MATERIAL
Coming soon ....
REFERENCES ()
CITATIONS ()
EXTERNAL LINKS
PlumX Metrics
RECOMMENDATIONS
FAIR ASSESSMENT
Coming soon ....
JUPYTER LAB
Coming soon ....