Mis-specified model supplementation

Patent metadata

Field	Value
Publication number	US-10885241-B2
Application number	US-201815861434-A
Country	US
Kind code	B2
Filing date	Jan 3, 2018
Priority date	Jan 3, 2018
Publication date	Jan 5, 2021
Grant date	Jan 5, 2021

Abstract

Methods and systems for generating output of a simulation model in a simulation system are described. In an example, a processor may retrieve observed output data from a memory. The observed output data may be generated based on a simulation operator of a simulation model. The processor may further optimize a generalization error of a distance measure between the observed output data and model output data. The model output data may be generated based on a high-fidelity operator. The processor may further determine a correction operator based on the optimized generalization error of the distance measure. The processor may further append the correction operator to the simulation operator to produce a supplemented operator. The processor may further generate supplemented output data by applying the simulation model with the supplemented operator on a set of inputs.

First claim

What is claimed is:

1. A method for generating output of a simulation model in a simulation system, the method comprising:
retrieving, by a processor, observed output data from a memory, wherein the observed output data is generated based on a simulation operator of the simulation model;
optimizing, by the processor, a generalization error of a distance measure between the observed output data and model output data, wherein the model output data is generated based on a high-fidelity operator;
determining, by the processor, a correction operator based on the optimized generalization error of the distance measure;
appending, by the processor, the correction operator to the simulation operator to produce a supplemented operator without modifying the simulation operator; and
generating, by the processor, supplemented output data by applying the simulation model with the supplemented operator on a set of inputs.

2. The method of claim 1 , wherein optimizing the generalization of the distance measure includes determining a set of solutions that minimizes the distance measure.

3. The method of claim 1 , wherein optimizing the generalization of the distance measure is based on a constraint to bound a rank of the correction operator within a value.

4. The method of claim 1 , wherein optimizing the generalization of the distance measure includes:
allocating, by the processor, a portion of the memory for the correction operator;
determining, by the processor, a set of gradients for each subset of the correction operator;
determining, by the processor, a set of eigenvectors associated with each set of gradients; and
populating, by the processor, the allocated portion of the memory with the determined set of eigenvectors.

5. The method of claim 4 , wherein determining the set of eigenvectors includes determining a set of largest eigenvectors of each set of gradients.

6. The method of claim 1 , wherein the distance measure is represented by a Frobenius norm function.

7. The method of claim 1 , wherein the simulation operator is a non-linear operator.

8. The method of claim 1 , wherein the observed output data is of a first fidelity, and the supplemented output data is of a second fidelity greater than the first fidelity.

9. A system effective to generate output of a simulation model in a simulation system, the system comprising:
a memory configured to store a simulation model, wherein the simulation model includes a simulation operator;
a simulation module configured to be in communication with the memory, the simulation module is configured to:
generate observed output data based on the simulation operator;
store the observed output data in the memory;
generate model output data based on a high-fidelity operator;
store the model output data in the memory;

a supplementation module configured to be in communication with the memory, the supplementation module is configured to:
retrieve the observed output data from the memory;
optimize a generalization error of a distance measure between the observed output data and model output data; and
determine a correction operator based on the optimized generalization error of the distance measure;

a processor configured to be in communication with the simulation module and the supplementation module, the processor is configured to append the correction operator to the simulation operator to produce a supplemented operator without modifying the simulation operator; and
the simulation module is further configured to generate supplemented output data by application of the simulation model with the supplemented operator on a set of inputs.

10. The system of claim 9 , wherein the supplementation module is further configured to determine a set of solutions that minimizes the distance measure in order to optimize the generalization error of the distance measure.

11. The system of claim 9 , wherein the supplementation module is further configured to bound a rank of the correction operator within a value in order to optimize the generalization error of the distance measure.

12. The system of claim 9 , wherein the supplementation module is further configured to:
allocate a portion of the memory for the correction operator;
determine a set of gradients for each subset of the correction operator;
determine a set of eigenvectors associated with each set of gradients; and
populate the allocated portion of the memory with the determined set of eigenvectors.

13. The system of claim 12 , wherein determination of the set of eigenvectors includes a determination a set of largest eigenvectors of each set of gradients.

14. The system of claim 9 , wherein the distance measure is represented by a Frobenius norm function.

15. The system of claim 9 , wherein the observed output data is of a first fidelity, and the supplemented output data is of a second fidelity greater than the first fidelity.

16. A computer program product for generating output of a simulation model in a simulation system, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor of a device to cause the device to:
retrieve observed output data from a memory, wherein the observed output data is generated based on a simulation operator of a simulation model;
optimize a generalization error of a distance measure between the observed output data and model output data, wherein the model output data is generated based on a high-fidelity operator;
determine a correction operator based on the optimized generalization error of the distance measure;
append the correction operator to the simulation operator to produce a supplemented operator without modifying the simulation operator; and
generate supplemented output data by applying the simulation model with the supplemented operator on a set of inputs.

17. The computer program product of claim 16 , wherein the program instructions are further executable by the device to cause the device to:
determine a set of solutions that minimizes the distance measure; and
bound a rank of the correction operator within a value in order to optimize the generalization error of the distance measure.

18. The computer program product of claim 16 , wherein the program instructions are further executable by the device to cause the device to:
allocate a portion of the memory for the correction operator;
determine a set of gradients for each subset of the correction operator;
determine a set of eigenvectors associated with each set of gradients; and
populate the allocated portion of the memory with the determined set of eigenvectors.

19. The computer program product of claim 16 , wherein the determination of the set of eigenvectors includes a determination of a set of largest eigenvectors of each set of gradients.

20. The computer program product of claim 16 , wherein the observed output data is of a first fidelity, and the supplemented output data is of a second fidelity greater than the first fidelity.

Assignees

• IBM
• Tufts College

Inventors

• Horesh Lior
• Hao Ning
• Horesh Raya
• Nahamoo David
• Kilmer Misha E

Classifications

• G06F30/20 Primary

  Design optimisation, verification or simulation (optimisation, verification or simulation of circuit designs G06F30/30) · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06F7/023

  adaptive, e.g. self learning · CPC title

• G06F2111/10

  Numerical modelling · CPC title