1
$\begingroup$

I have few questions about Deep Q Network.

  1. Does DQN only accept image frames as input? I have never hear (read) a paperwork where it doesn't use image frames.

  2. If the first is a No, then does image frames as input for training faster than any other options? For example, in any atari game, is frames input faster to train while having the same perfomance quality?

I know an image frames is but numbers too, like other datas, but from what I know even a low quality frame is considered huge.

$\endgroup$
1
$\begingroup$

DQNs don't only accept image frames as input. For instance, this DQN for the CartPole environment takes in a state with only 4 elements (the position and velocity of the cart and the angle and velocity of the pole.)

Images aren't necessarily more efficient representations than other options. For instance, here is another DQN for the same CartPole environment which uses an image representation and is much less efficient. A couple reasons people often use image representations include:

  1. It's the only available representation for the task.
  2. Other models uses image representations and you want to compare.
  3. You want to pass the network the same information available to humans.

If none of these apply, though, if you have a representation which gives you more insight into the position or trajectory of elements in the screen it's probably best to use that rather than a raw image.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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