Causal Machine Learning Surrogate for Simulator Cost Reduction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Physics-based simulators require significant computing resources and time, making it costly and environmentally intensive to evaluate multiple configurations of real-world systems, especially when searching for optimal input combinations.

Innovation Solution

A machine learning model is trained with data generated by a simulator to approximate its behavior, utilizing domain knowledge to identify non-interacting input variables and reduce the number of simulator iterations needed, allowing for more efficient training and prediction of system configurations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a physics-based simulator is used to evaluate configurations of a real-world system, then the accuracy of the evaluation is improved, but the computing cost and time required increase significantly

Engineering Contradiction:

Improveevaluation accuracyVSAvoidcomputing cost

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent creates a surrogate model that copies the essential behavior of the physics-based simulator. This surrogate model is trained on a subset of simulation data and can predict outcomes without requiring full simulator execution, thus maintaining evaluation accuracy while dramatically reducing computing costs and energy consumption.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent uses a computationally inexpensive surrogate model to replace the expensive physics-based simulator for most evaluation tasks. The surrogate model, being cheaper and faster to execute, can be used repeatedly without the high computational overhead of the original simulator, effectively treating the expensive simulator as a resource to be minimized rather than continuously deployed.

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

2Measurement precision

If a physics-based simulator is used to evaluate configurations of a real-world system, then the accuracy of the evaluation is improved, but the time required increases significantly

Engineering Contradiction:

Improveevaluation accuracyVSAvoidevaluation time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent performs preliminary training of the surrogate model using a subset of simulation data before actual evaluation. By pre-training the model on representative data, the system can quickly answer new evaluation questions without re-executing the full simulator, significantly reducing the time required for repeated configuration evaluations.

Inventive Principle:

Principle #10Preliminary action

3Reliability

If the simulator is run multiple times to search for optimal input combinations, then the quality of the decision-making is improved, but the computing cost and environmental impact increase

Engineering Contradiction:

Improvedecision-making qualityVSAvoidenvironmental impact

Core Design Contradiction:

ReliabilityVSObject-generated harmful factors

Solution Approach 1:

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent changes the computational parameters by switching from executing the full physics-based simulator to using a trained surrogate model. This parameter change allows for multiple iterations of evaluation and optimization searches without proportionally increasing energy consumption, as the surrogate model requires minimal computational resources compared to the original simulator.

Inventive Principle:

Principle #35Parameter changes

4Productivity

If domain knowledge is used to identify non-interacting input variables, then the number of simulator iterations is reduced, but the complexity of data selection and model training increases

Engineering Contradiction:

Improvenumber of iterationsVSAvoiddata selection complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent segments the input variables into non-interacting groups based on domain knowledge. By identifying which variables can be varied independently without affecting each other's impact on the output, the system can strategically select a reduced set of iterations that still covers the essential parameter space, reducing total simulator calls while managing the complexity of informed data selection.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS20230162062A1Optimization and decision-making using causal aware machine learning models trained from simulators

Publication Date: 2023.05.25 MICROSOFT TECHNOLOGY LICENSING LLC

AI summary

Techniques are described herein for reducing the computing cost of decision-making when simulating a real-world system. A machine learning model is trained using data generated by a simulator of the real-world system. Knowledge about how the simulator is implemented is used to improve the efficiency of the machine learning model and to improve the relevance of data selected to train the machine learning model. For example, structural knowledge—the flow of input variables through components of the simulator—is used to determine a causal relationship between input variables. Having identified the causal relationship, the number of simulator iterations used to generate training data may be reduced. Furthermore, large complex machine learning models may be replaced with smaller, more efficient models. Additionally, or alternatively, causal relationships between input variables are identified during training, enabling further refinement of input selection and model design.