Learning Attentional Communication with a Common Network for Multiagent Reinforcement Learning

Figure 2

Extracting significant information from the common network based on the multihead attention mechanism, where means is a linear layer and and are also linear layers.