4
$\begingroup$

One of the challenges that arise in reinforcement learning, and not in other kinds of learning, is the trade-off between exploration and exploitation. Single-step reinforcement learning model is original of k-armed bandit.
But there are two kind of k-armed bandit problem: a stationary and non-stationary k-armed bandit problem, What is the difference between them?

$\endgroup$

2 Answers 2

2
$\begingroup$

According to wikipedia, in a non-stationary multi-armed bandit problem the "underlying model can change during play."

$\endgroup$
1
$\begingroup$

In case of stationary bandit problems, the same bandit and distributions from which rewards are being sampled at each time step stays the same. This is called a stationary problem. To explain it with another example, say you get a reward of 1 every time a coin is tossed, and the result is head. Say after 1000 coin tosses due to wear and tear the coin becomes biased then this will become a non-stationary problem.

$\endgroup$

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.