Social-Aware Caching Strategy Based on Joint Action Deep Reinforcement Learning
Algorithm 1
Caching strategy based on a JADQN.
Input: inactive users remaining space: , transmission distance of users: , delay threshold: , the parameter of main network: , the parameter of target network: and read the data from dataset.
Output: the cache hit ratio .
Initialize:, memory size: , leaning rate , explore rate: , discount rate: , the number of iterations:, the number of minibatches:.
ifthen
for episode=1do
Update ;
for t =do
Random generation probability:
if:
Randomly generated action:
else:
Obtained a joint action by according to
end if
Observe
Store the transactionin replay memory , uniformly sample minibatches from
Optimize error between -network and learning targets, using variant of stochastic gradient descent
Each step updates the parameters of target -network
