1
$\begingroup$

The creation of negative rewards leads to the chance of Q-values being negative. However, networks with relu or sigmoid activations, just cannot predict negative values. This will lead to a case where erroneous Q-values are being predicted. Is my understanding correct?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
2
$\begingroup$

A network with ReLU activation can predict negative values; we put ReLU between the hidden layers but return the output of the final layer without any activation function, or with a linear activation function to scale the output.

Improve this answer
$\endgroup$
1
  • 3
    $\begingroup$ It is worth mentioning that some kind of non-linear activation function is necessary for hidden layers to be useful. Only in the output layer should be made linear in most cases (I hesitate to say all because although I cannot think of any reason, someone will probably know of a case where it is useful to have a linear hidden layer). $\endgroup$
    – Neil Slater
    Jun 10 at 7:17

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .