Moving Y target?

Course Q&A Machine Learning Specialization Unsupervised Learning, Recommenders, Reinforcement
1

Could someone help me understand this part?
Why is Y a moving target? From my understanding Y’ is also a moving target since the weights apply to both of them.

And how is this solved by the Target Q network?

image
image2011×658 42.8 KB

1 Like
2

Hi @jjdniz

Both Y and Y′ are moving targets because they depend on the current parameters of the Q-network, which are updated during training. In training, the target Q-network uses a separate network with frozen parameters to estimate Y'.