TECHNIQUES FOR MODELING BEHAVIORS OF SYSTEMS VIA TRANSFORMATIONS OF AUTHORITATIVE MODELS

• US 20170154139A1
• Filed: 11/11/2016
• Published: 06/01/2017
• Est. Priority Date: 11/30/2015
• Status: Active Grant

• Current Assignee: Autodesk, Inc.
• Original Assignee: Autodesk, Inc.
• Inventors: IORIO, Francesco, HASHEMI, Ali BARADARAN
• Granted Patent: US 11,347,904 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 30/20 Design optimisation, verification or simulation optimisation, verification or simulation of circuit designs G06F30/30
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]

Abstract

In one embodiment, a model generator generates a new model for a behavior of a system based on an existing, authoritative model. First, a mapping generator generates a mapping model that maps authoritative values obtained via the authoritative model to measured values that represent the behavior of the system. Subsequently, the model generator creates the new model based on the authoritative model and the mapping model. In this fashion, the mapping model indirectly transforms the authoritative model to the new model based on the measured values. Advantageously, the authoritative model enables the model generator to increase a rate of accuracy improvement experienced while developing the new model compared to a rate of accuracy improvement that would be experienced were the new model to be generated based on conventional modeling techniques. In particular, for a given sampling budget, the model generator improves the accuracy of the new model.

Claims

1. A computer-implemented method for generating a model for a behavior of a system, the method comprising:

generating a first sample data that includes a first authoritative value that is obtained from an authoritative model based on a first combination of system parameters and a first measured value that represents the behavior of the system at the first combination of system parameters;

training an untrained mapping model based on the first sample data to generate a partially trained mapping model;

training the partially trained mapping model based on at least one other sample data to generate a trained mapping model, wherein the trained mapping model maps values obtained from the authoritative model to values that represent the behavior of the system at different combinations of system parameters; and

generating the model for the behavior of the system based on the trained mapping model and the authoritative model, wherein the model estimates the behavior of the system for different combinations of system parameters.

2. The computer-implemented method of claim 1, wherein the first trained mapping model implements a support vector machine, a Gaussian process, a polynomial regression, or a piecewise linear regression modeling method.

3. The computer-implemented method of claim 1, wherein the authoritative model comprises at least one of a function and a plurality of authoritative values.

4. The computer-implemented method of claim 1, wherein the trained mapping model implements a first modeling method, and generating the model for the behavior of the system comprises:

computing a first accuracy that is associated with the first modeling method based on an evaluation model that implements the first modeling method;

if the first accuracy is higher than a second accuracy that is associated with a second modeling method, then composing the trained mapping model with the authoritative model, or

if the first accuracy is not higher than the second accuracy, then composing a different trained mapping model that is associated with the second modeling method with the authoritative model.

5. The computer-implemented method of claim 4, wherein computing the first accuracy comprises performing k-fold cross validation operations on the evaluation model.

6. The computer-implemented method of claim 4, further comprising computing the second accuracy substantially in parallel to computing the first accuracy.

7. The computer-implemented method of claim 1, further comprising selecting the authoritative model from a plurality of existing models based on a similarity between the authoritative model and the behavior of the system.

8. The computer-implemented method of claim 1, wherein the system comprises a computer system, a database system, or a design optimization system.

9. The computer-implemented method of claim 1, wherein training the partially trained mapping model comprises:

selecting a second combination of system parameters;

generating a second sample data that includes a second authoritative value that is obtained from an authoritative model based on the second combination of system parameters and a second measured value that represents the behavior of the system at the second combination of system parameters; and

training the partially trained mapping model based on the second sample data.

10. The computer-implemented method of claim 9, wherein selecting the second combination of system parameters comprises determining a combination of system parameters that corresponds to a maximum estimated uncertainty that is associated with the partially trained mapping model.

11. A computer-readable storage medium including instructions that, when executed by a processing unit, cause the processing unit to generate a model for a behavior of a system by performing the steps of:

generating a first sample data that includes a first authoritative value that is obtained from an authoritative model based on a first combination of system parameters and a first measured value that represents the behavior of the system at the first combination of system parameters;

training an untrained mapping model based on the first sample data to generate a partially trained mapping model;

training the partially trained mapping model based on at least one other sample data to generate a trained mapping model, wherein the trained mapping model maps values obtained from the authoritative model to values that represent the behavior of the system at different combinations of system parameters; and

generating the model for the behavior of the system based on the trained mapping model and the authoritative model, wherein the model estimates the behavior of the system for different combinations of system parameters.

12. The computer-readable storage medium of claim 11, wherein the first trained mapping model implements a support vector machine, a Gaussian process, a polynomial regression, or a piecewise linear regression modeling method.

13. The computer-readable storage medium of claim 11, wherein the authoritative model comprises at least one of a function and a plurality of authoritative values.

14. The computer-readable storage medium of claim 11, wherein the trained mapping model implements a first modeling method, and generating the model for the behavior of the system comprises:

computing a first accuracy that is associated with the first modeling method based on an evaluation model that implements the first modeling method;

if the first accuracy is higher than a second accuracy that is associated with a second modeling method, then composing the trained mapping model with the authoritative model, or

if the first accuracy is not higher than the second accuracy, then composing a different trained mapping model that is associated with the second modeling method with the authoritative model.

15. The computer-readable storage medium of claim 11, wherein generating the model for the behavior of the system comprises composing the mapping model with the authoritative model.

16. The computer-readable storage medium of claim 11, wherein the authoritative model is associated with an authoritative behavior that is structurally similar to the behavior of the system.

17. The computer-readable storage medium of claim 11, wherein the system comprises a computer system, a database system, or a design optimization system.

18. The computer-readable storage medium of claim 11, wherein the first measured value comprises a scalar or a vector.

19. The computer-readable storage medium of claim 11, wherein training the partially trained mapping model comprises:

selecting a second combination of system parameters;

generating a second sample data that includes a second authoritative value that is obtained from an authoritative model based on the second combination of system parameters and a second measured value that represents the behavior of the system at the second combination of system parameters; and

training the partially trained mapping model based on the second sample data.

20. A system comprising:

a memory storing a model generator; and

a processor that is coupled to the memory and, when executing the model generator, is configured to;

generate a first sample data that includes a first authoritative value that is obtained from an authoritative model based on a first combination of system parameters and a first measured value that represents a behavior of a system at the first combination of system parameters;

train an untrained mapping model based on the first sample data to generate a partially trained mapping model;

train the partially trained mapping model based on at least one other sample data to generate a trained mapping model, wherein the trained mapping model maps values obtained from the authoritative model to values that represent the behavior of the system at different combinations of system parameters; and

generate the model for the behavior of the system based on the trained mapping model and the authoritative model, wherein the model estimates the behavior of the system for different combinations of system parameters.