Performance Evaluation of Multiagent Reinforcement Learning Based Training Methods for Swarm Fighting

Table 5

Hyperparameters used for this experiment.

Type		MARL	MARL-BC
Hyperparameters	Batch size	1024	1024
	Buffer size	20480	20480
	Learning rate	0.0001	0.0001
	Entropy bonus	0.005	0.005
	Num epoch	3	3
Network settings	Hidden units	512	512
	Num layers	3	3
Reward signals	Discount factor	0.99	0.99
	Strength	1.0	1.0
Behavior cloning	Steps	/	100 M
	Strength	0.5	0.5