No label defined (Q4727232)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 4001250
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | No label defined |
scientific article; zbMATH DE number 4001250 |
Statements
1987
0 references
conjugate prior
0 references
sequential design
0 references
two-armed bandit
0 references
switching cost
0 references
total expected discounted reward
0 references
critical value function
0 references
0.9138286
0 references
0.8785167
0 references
0.8604362
0 references
Suppose that a sequence of N independent trials is performed in one of two experiments \(e_ 0\) and \(e_ 1\). The outcome by performing \(e_ i\) is a value \(z_ i\) of a nonnegative random variable \(Z_ i\) with probability density function \(f(\cdot;u_ i)\) depending on the parameter \(u_ i\), \(i=0,1\). It is assumed that one of these parameters, \(u_ 1\), is known and \(u_ 0\) has a prior density function g(\(\cdot;x,y)\) with parameters x and y. If \(z_ i\) occurs, then the reward r is \(az_ i\) with \(a=(EZ_ 1)^{-1}\); at the next step r is discounted by a factor \(\beta\), \(0<\beta \leq 1\), \(i=0,1\). If the same experiment is performed on the next trial, no cost is incurred. By switching from \(e_ i\) to \(e_{1-i}\) on the next trial a cost \(c_{i,1-i}\), \(i=0,1\), is incured. If \(e_ i\) is performed on the last trial, the final reward obtained is \(b_ i\), \(i=0,1\). The goal is to maximize the total expected discounted reward. In this paper several properties of the critical value function which specifies the optimal strategy are derived. Maximum expected rewards under special conditions are also given, by using the critical value function where values can be numerically computed.
0 references