October 2018
Beginner
362 pages
9h 32m
English
Let's look back at the Q- function optimization process:

You might notice that this function has a unique property in that it's recursive; one set of Q-values depend on the value of the other set of Q-values. This becomes a problem in training; if we change one set of values, we'll end up changing the other set of values. To put it simply, as we get closer to the targeted Q-value, that Q-value moves even further away. It is continually moving the finish line when you are about to finish a race!
To remedy this, the Q-network creates a copy of itself every 10,000 iterations, called a target network, which will represent ...
Read now
Unlock full access