A Entity wise Input

Neural Information Processing Systems 

It's critical to deal with the dynamic team composition in real-world multi-agent scenarios. MHA can easily achieve agents' partial observability with entity-wise expression. We commit our code in https://github.com/thu-rllab/SOG . 1 D Derivation of Message Summarizer For StarCraft Micromanagement Tasks, it takes about 20 hours to run one experiment. We summarize some hyper-parameters in Table. The radius of the home and the resource location & agent is 0.1 and 0.05, respectively.

Similar Docs  Excel Report  more

TitleSimilaritySource
None found