0
$\begingroup$

In section 2 of the paper Playing Atari with Deep Reinforcement Learning, it says a Q-network can be trained by minimizing a sequence of loss functions. I do not understand why there is more than one loss function in this case.

Loss functions for DQN

$\endgroup$

1 Answer 1

0
$\begingroup$

This sequence is really just a notation for iterative updates. It basically says, for the $i$th update, use your neural net parameters at timstep $i$ to make your prediction and use your neural net parameters from timestep $i-1$ to compute the target.

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.