Systems, methods, and computer program products for predictive accuracy for strategic decision support

• US 8,620,852 B1
• Filed: 12/16/2010
• Issued: 12/31/2013
• Est. Priority Date: 12/16/2010
• Status: Active Grant

• Current Assignee: The Boeing Co.
• Original Assignee: The Boeing Co.
• Inventors: Kipersztok, Oscar, Nodelman, Uri
• Primary Examiner(s): Gaffin, Jeffrey A
• Assistant Examiner(s): OLUDE AFOLABI, OLATOYOSI
• Application Number:
  US12/969,711
• Time in Patent Office: 1,111 Days
• Field of Search: 706/46, 706/50
• US Class Current: 706/52
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

Provided are improved systems, methods, and computer programs to facilitate predictive accuracy for strategic decision support using Continuous Time Bayesian Networks. A simple semantic is provided for both probabilistic and temporal relationships that can be automatically translated into functioning Bayesian and Continuous Time Bayesian Network formalisms and applied by a user for reasoning analysis, such as to assess a hypothesis or respond to a query with a probabilistic and temporal prediction of a future event.

Claims

1. A method, comprising:

providing an unconstrained domain model defined by domain concepts, probabilistic causal relationships between the domain concepts, and a temporal relationship between at least one pair of the domain concepts, wherein each probabilistic causal relationship includes a value for the weight of causal belief for the probabilistic causal relationship, and wherein each temporal relationship includes a value for the expected time to the effect for the temporal relationship;

receiving the probabilistic causal relationships between the domain concepts for the domain model from a user;

receiving the temporal relationships between the at least one pair of domain concepts for the domain from the user; and

transforming, using a processor, the unconstrained domain model into a Continuous Time Bayesian Network formalism for the domain model using at least the domain concepts and temporal relationships.

2. The method of claim 1, further comprising transforming, using the processor, the unconstrained domain model into a Bayesian network formalism using at least the domain concepts and the probabilistic causal relationships.

3. The method of claim 2, further comprising:

performing reasoning analysis according to a hypothesis or a query using the Bayesian network formalism and the Continuous Time Bayesian Network formalism for the domain model; and

obtaining a prediction of the probability of the timing for an event based upon the hypothesis or query.

4. The method of claim 1, wherein the domain model comprises at least two temporal relationships, wherein at least one temporal relationship is defined by a value for the expected time to the effect representing an instant effect, and wherein at least one temporal relationship is defined by a value for the expected time to the effect representing a specific period of time greater than an instant effect.

5. The method of claim 1, wherein transforming the unconstrained domain model into a Continuous Time Bayesian Network formalism includes creating a conditional intensity matrix for each temporal relationship.

6. The method of claim 5, wherein the conditional intensity matrix for at least one of the temporal relationships defines a binary-valued child variable with a single binary-valued parent relationship.

7. The method of claim 5, wherein the conditional intensity matrix for at least one of the temporal relationships correlates to the pair of the domain concepts X and Y with a positive probabilistic causal weight value p and expected time to effect T, such that when X=x₁, then Y transitions from y₀ to y₁ with probability p and expected time T and when X=x₀, then Y transitions from y₁ to y₀ with probability p and expected time T.

8. The method of claim 5, wherein the conditional intensity matrix for at least one of the temporal relationships correlates to the pair of the domain concepts X and Y with a negative probabilistic causal weight value p and expected time to effect T, such that when X=x₁, then Y transitions from y₁ to y₀ with probability p and expected time T and when X=x₀, then Y transitions from y₀ to y₁ with probability p and expected time T.

9. The method of claim 5, wherein at least a first domain concept has at least two parents having temporal relationships, and wherein transforming the unconstrained domain model into a Continuous Time Bayesian Network formalism includes creating an amalgamated single, conditional intensity matrix for the first domain concept from a matrix addition of all of the temporal relationships of the first domain concept with its parents having temporal relationships.

10. The method of claim 5, wherein a child domain concept Y having a plurality of parent domain concepts X with a positive probabilistic causal weight value p and expected time to effect T has a single, complete conditional intensity matrix for the temporal relationships between child domain concept and its parent domain concepts X, such that an increasing or occurring domain concepts X leads to an increasing or occurring domain concept Y with probability p over expected time T and a decreasing or not-occurring domain concepts X leads to a decreasing or not-occurring domain concept Y with probability p over expected time T.

11. The method of claim 5, wherein a child domain concept Y having a plurality of parent domain concepts X with a negative probabilistic causal weight value p and expected time to effect T has a single, complete conditional intensity matrix for the temporal relationships between child domain concept and its parent domain concepts X, such that an increasing or occurring domain concepts X leads to a decreasing or not-occurring domain concept Y with probability p over expected time T and a decreasing or not-occurring domain concepts X leads to an increasing or occurring domain concept Y with probability p over expected time T.

12. A system, comprising:

an unconstrained domain model defined by domain concepts, probabilistic causal relationships between the domain concepts, and a temporal relationship between at least one pair of the domain concepts, wherein each probabilistic causal relationship includes a value for the weight of causal belief for the probabilistic causal relationship, and wherein each temporal relationship includes a value for the expected time to the effect for the temporal relationship;

a reasoning tool configured to perform reasoning analysis according to a hypothesis or a query using the Bayesian network formalism and the Continuous Time Bayesian Network formalism for the domain model;

an output tool configured to output a prediction of the probability of the timing for an event based upon the hypothesis or query; and

a processor configured to transform the unconstrained domain model into a Continuous Time Bayesian Network formalism for the domain model using at least the domain concepts and temporal relationships.

13. The system of claim 12, wherein the processor is further configured to transform the unconstrained domain model into a Bayesian network formalism using at least the domain concepts and the probabilistic causal relationships.

14. The system of claim 12, wherein the processor is further configured to transform the unconstrained domain model into the Continuous Time Bayesian Network formalism by creating a conditional intensity matrix for each temporal relationship.

15. A computer program comprising a computer-useable memory having control logic stored therein for facilitating predictive accuracy for strategic decision support, the control logic comprising:

a first code configured to provide an unconstrained domain model defined by domain concepts, probabilistic causal relationships between the domain concepts, and a temporal relationship between at least one pair of the domain concepts, wherein each probabilistic causal relationship includes a value for the weight of causal belief for the probabilistic causal relationship, and wherein each temporal relationship includes a value for the expected time to the effect for the temporal relationship; and

a second code configured to transform the unconstrained domain model into a Continuous Time Bayesian Network formalism for the domain model using at least the domain concepts and temporal relationships, wherein the second code is further configured to transform the unconstrained domain model into a Continuous Time Bayesian Network formalism by creating a conditional intensity matrix for each temporal relationship.

16. The computer program of claim 15, wherein the second code is further configured to transform the unconstrained domain model into a Bayesian network formalism using at least the domain concepts and the probabilistic causal relationships.

17. The computer program of claim 16, further comprising:

third code configured to perform reasoning analysis according to a hypothesis or a query using the Bayesian network formalism and the Continuous Time Bayesian Network formalism for the domain model; and

fourth code configured to obtain a prediction of the probability of the timing for an event based upon the hypothesis or query.