scientific article; zbMATH DE number 3854141
[For the entire collection see Zbl 0517.00013.] Let \(X_{j1},X_{j2},...\); \(j=1,...,k\) be k stochastic processes each of which consists of identically distributed random variables. Suppose that the probability distribution of the random variables of the jth process depend on a parameter \(\vartheta_ j\) and that at least one of these parameters is unknown. At each time one stochastic process (one arm) is selected and observed. The mth observation is weighted by a factor \(\alpha_ m\) and the problem is to find a strategy maximizing the expected weighted sum over the finite or infinite horizon. This paper deals with the problem where the weighting (discounting) factors are also random variables. Various special cases are discussed and illustrated by revealing numerical examples such as where the discount sequence is not completely known or is not observable or is a mixture of nonrandom sequences. The question when are problems with random discount factors equivalent to certain nonrandom discount factor problems is also discussed.
- A penalized bandit algorithm
- Optimal, recursive procedures of identification
- Order scoring, bandit learning and order cancellations
- Batched bandit problems
- Negatively Correlated Bandits
- On the problem of the two-armed bandit with impulse controls and discounting
- Two-Armed Bandit Strategies that Discount Past and Future
- Poisson Version of the Two-Armed Bandit Problem with Discounting
- A doscounted uniform one-armed bandit problem
- Bandits and Experts in Metric Spaces
- On the Worth of Perfect Information in Bandits with Random Discounting
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3322965)