Globally convergent system and method for automated model discovery

Patent metadata

Field	Value
Publication number	US-10719637-B2
Application number	US-201514755942-A
Country	US
Kind code	B2
Filing date	Jun 30, 2015
Priority date	Jun 30, 2015
Publication date	Jul 21, 2020
Grant date	Jul 21, 2020

Abstract

Methods and systems for model discovery include forming a mathematical program based on a set of observational data to generate an objective function and one or more constraints. The mathematical program represents a model space as an expression tree comprising operators and operands. The mathematical program is solved by optimizing the objective function subject to the one or more constraints to determine a model in the model space that best fits the set of observational data.

First claim

The invention claimed is:

1. A method for performing predictions with automatically discovered models, comprising:
forming a mathematical program based on a set of observational data from a physical system to generate an objective function Σ s∈V α s +δ, where s is a node in a set of possible nodes V, α s is a binary parameter that represents whether the node s is present in an expression tree, and δ is a measure of model complexity that includes a maximum number of operators in a model and a respective maximum number for each operator in the model that represents a number of times that the operator can appear in the model, wherein the objective function captures a fidelity of model predictions to observations, and to generate one or more constraints, wherein the mathematical program represents a model space as the expression tree comprising operators and operands, wherein the operators include variables, coefficients, and mathematical operations and the constraints include numerical constraints, which assign correct values of nodes of the expression tree to produce correct values for each input, structural constraints, which enforce rules regarding the structure of the expression tree, and mixed constraints, represented by both continuous and integer variables;
solving the mathematical program by minimizing the objective function, subject to the one or more constraints, to determine the model in the model space, including a functional form and parameters of the model with a minimal complexity, that best fits the set of observational data; and
predicting future behavior of the physical system using the determined model.

2. A computer readable storage medium comprising a computer readable program for performing predictions with automatically discovered models, wherein the computer readable program when executed on a computer causes the computer to perform the steps of:
forming a mathematical program based on a set of observational data from a physical system to generate an objective function Σ s∈V α s +δ, where s is a node in a set of possible nodes V, α s is a binary parameter that represents whether the node s is present in an expression tree, and δ is a measure of model complexity that includes a maximum number of operators in a model and a respective maximum number for each operator in the model that represents a number of times that the operator can appear in the model, wherein the objective function captures a fidelity of model predictions to observations, and to generate one or more constraints, wherein the mathematical program represents a model space as the expression tree comprising operators and operands, and wherein the operators include variables, coefficients, and mathematical operations and the constraints include numerical constraints, which assign correct values of nodes of the expression tree to produce correct values for each input, structural constraints, which enforce rules regarding the structure of the expression tree, and mixed constraints, represented by both continuous and integer variables;
solving the mathematical program by minimizing the objective function, subject to the one or more constraints, to determine the model in the model space, including a functional form and parameters of the model with a minimal complexity, that best fits the set of observational data; and
predicting future behavior of the physical system using the determined model.

3. A system for performing predictions with automatically discovered models, comprising:
a mathematical program generation module comprising a processor configured to form a mathematical program based on a set of observational data from a physical system to generate an objective function Σ s∈V α s +δ, where s is a node in a set of possible nodes V, α s is a binary parameter that represents whether the node s is present in an expression tree, and δ is a measure of model complexity that includes a maximum number of operators in model and a respective maximum number for each operator in the model that represents a number of times that the operator can appear in the model, wherein the objective function captures a fidelity of model predictions to observations, and to generate one or more constraints, wherein the mathematical program represents a model space as the expression tree comprising operators and operands, and wherein the operators include variables, coefficients, and mathematical operations and the constraints include numerical constraints, which assign correct values of nodes of the expression tree to produce correct values for each input, structural constraints, which enforce rules regarding the structure of the expression tree, and mixed constraints, represented by both continuous and integer variables;
a solver configured to solve the mathematical program by minimizing the objective function subject to the one or more constraints to determine the model in the model space, including a functional form and parameters of the model with a minimal complexity, that best fits the set of observational data and to predict future behavior of the physical system using the determined model.

Assignees

• IBM

Inventors

• Avron Haim
• Horesh Lior
• Liberti Leo S
• Nahamoo David

Classifications

• G06F17/10 Primary

  Complex mathematical operations {(function generation by table look-up G06F1/03; evaluation of elementary functions by calculation G06F7/544)} · CPC title

• G06F30/20 Primary

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