Decision engine and method and applications thereof

• US 20050246306A1
• Filed: 05/16/2005
• Published: 11/03/2005
• Est. Priority Date: 07/18/2000
• Status: Abandoned Application

• Current Assignee: Jeremy Link, Lincoln T. Evans-Beauchamp
• Original Assignee: Jeremy Link, Lincoln T. Evans-Beauchamp
• Inventors: Evans-Beauchamp, Lincoln T., Link, Jeremy
• Application Number:
  US11/130,525
• Publication Number:
  US 20050246306A1
• Time in Patent Office: Days
• US Class Current: 706/50
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

A method of creating a decision engine including a Bayesian network. The method includes retrieving data from a client database and forming a focus database; applying a set of initial rules to the focus database to form at least two nodes; applying a first learning process to determine a set of arcs to be applied between the at least two nodes; applying a second learning process to determine a set of states to be applied within each node; applying a third learning process to determine a set of probabilities applicable to the states learned in the second learning process; and applying a fourth learning process to update a structure of the at least two nodes, the set of arcs, the set of states within each node, and the set of probabilities for the states.

Claims

1. A method of creating a programmable decision engine for jury selection in a computer-readable medium including a Bayesian network, comprising:

retrieving data from a client database, said client database containing demographic information about a jury pool and forming a first database of demographic information about a jury pool, wherein the retrieving includes retrievable data from a customer database, said customer database containing demographic information about a jury pool, and retrieving data from a data stream;

applying a set of initial rules to the first database to form at least two nodes relating to variables in the demographic information;

applying a first learning process to determine a set of arcs to be applied between the at least two nodes;

applying a second learning process to determine a set of states to be applied within each node, the set of states relating to values taken by the variables;

applying a third learning process to determine a set of probabilities applicable to the states learned in the second learning process; and

applying a fourth learning process to update a structure of the at least two nodes, the set of arcs, the set of states within each node, and the set of probabilities for the states, such that the first database of jury pool demographic information is updated and contains updated probabilities for the states relating to each node, and further contains updated information relating to the arcs between the nodes.

2. The method of claim 1, wherein the first learning process includes parameter learning.

3. The method of claim 1, wherein the second learning process includes state learning.

4. The method of claim 1, wherein the third learning process includes parameter learning.

5. The method of claim 1, wherein the fourth learning process includes structural learning.

6. The method of claim 1, wherein the client database is a relational database.

7. The method of claim 1, further comprising creating, accessing, and modifying an AD tree.

8. The method of claim 1, further comprising pre-analyzing the customer database to create a data management system.

9. The method of claim 1, wherein said forming includes counting the occurrences of possible combinations of data in the client database, and determining the frequencies of the data.

10. The method of claim 1, further comprising performing prior discretization of data in the client database to lower noise in the data.

11. The method of claim 1, further comprising applying expert knowledge to data in the first database.

12. The method of claim 3, wherein the applying a state learning includes applying a clustering algorithm.

13. The method of claim 5, wherein the applying a structural learning includes applying a process selected from one of the set consisting of:

directed Pareto, naï

ve Bayesian, directed Bayesian, recursive Pareto, whole Pareto, single MDL, multiple MDL, recursive naï

ve Bayesian, and whole Bayesian.

14. The method of claim 1, wherein the initial rules include a rule that columns within the client database correspond to the at least two nodes.

15. A method of using a programmable decision engine for jury selection in a computer-readable medium including a Bayesian network, comprising:

retrieving data from a client database, said client database containing demographic information about a jury pool and forming a first database of demographic information about a jury pool, wherein the retrieving includes retrievable data from a static customer database, said customer database containing demographic information about a jury pool, and retrieving data from a data stream;

applying a set of initial rules to the first database to form at least two nodes relating to variables in the demographic information;

applying a first learning process to determine a set of arcs to be applied between the at least two nodes;

applying a second learning process to determine a set of states to be applied within each node, the set of states relating to values taken by the variables;

applying a third learning process to determine a set of probabilities applicable to the states learned in the second learning process; and

applying a fourth learning process to update a structure of the at least two nodes, the set of arcs, the set of states within each node, and the set of probabilities for the states;

applying evidence to at least one of the nodes; and

updating the structure according to the applied evidence using at least one of the first, second, third, or fourth learning processes, such that the first database of jury pool demographic information is updated and contains updated probabilities for the states relating to each node, and further contains updated information relating to the arcs between the nodes.

16. The method of claim 15, further comprising displaying at least one of the set of probabilities applicable to the states in at least one of the nodes.

17. The method of claim 15, further comprising creating, accessing, and modifying a decision tree.

18. The method of claim 17, wherein a target of the modifying is determined using an intelligent decision analysis algorithm.

19. A computer program, residing on a computer-readable medium, for creating and using a programmable decision engine for jury selection including a Bayesian network, the computer program comprising instructions for causing a computer to:

retrieve data from a client database, said client database containing demographic information about a jury pool and forming a first database of demographic information about a jury pool, wherein the retrieval includes a retrieval of data from a static customer database, said customer database containing demographic information about a jury pool, and a retrieval of data from a data stream;

apply a set of initial rules to the first database to form at least two nodes relating to variables in the demographic information;

apply a first learning process to determine a set of arcs to be applied between the at least two nodes;

apply a second learning process to determine a set of states to be applied within each node, the set of states relating to values taken by the variables;

apply a third learning process to determine a set of probabilities applicable to the states learned in the second learning process; and

apply a fourth learning process to update a structure of the at least two nodes, the set of arcs, the set of states within each node, and the set of probabilities for the states, such that the first database of jury pool demographic information is updated and contains updated probabilities for the states relating to each node, and further contains updated information relating to the arcs between the nodes.