We normally use the online network for action selection and the target network for evaluation , would there be a difference if we switched the roles?

Because in the case Of Double Q learning, we alternate between the two estimators, this isn’t the case in Double deep Q learning

If our target network is the one that actually overestimated it’s values how do we correct that

Eg Q(s,a) =R + y Q(s’, argmax(Q*(s’,a,w-)),w)

In this example our online network has weights w And our target network has weights w- As you can see the target network performs action selection And our online network performs evaluation

Would there be any issue with doing it this way?

  • $\begingroup$ Hi @Chuki Bom, any chance you edit your post to provide more details on the context of the problem and can you also make your question more explicit as to what you are specifically asking? $\endgroup$
    – shepan6
    Commented Aug 3, 2020 at 13:57
  • $\begingroup$ Done that, added necessary details $\endgroup$
    – Chuki Bom
    Commented Aug 3, 2020 at 19:43


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.