Active Learning Input Selection for Uncertainty Quantification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for inferring statistics of interest in predicting component failures, such as turbine failure, require expensive experiments and computer simulations, which are computationally intensive and affected by uncertainties in input parameters like manufacturing tolerances and noise in measurements.
Innovation Solution
A system and method for intelligent selection of inputs using a machine learning model to maximize information gain, incorporating input uncertainty and estimating metrics of interest, allowing for efficient allocation of computational resources and quantifying uncertainties to achieve accurate predictions with less costly data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional measuring systems are used to perform estimation, then measurement precision is improved, but loss of energy increases due to expensive experiments and large amounts of information required
Solution Approach 1:
The patent creates a surrogate model (copy) of the expensive computational model that can be evaluated rapidly. This surrogate model is trained on a limited set of expensive model evaluations and then used to make predictions without requiring additional expensive computations, thus reducing energy loss while maintaining estimation accuracy.
Solution Approach 2:
The patent replaces expensive, computationally intensive model evaluations with cheap, rapid surrogate model predictions. The surrogate model acts as a disposable approximation that can be queried many times without the high computational cost of the original model, significantly reducing energy consumption.
2Measurement precision
If conventional measuring systems are used to perform estimation, then measurement precision is improved, but loss of time increases due to expensive experiments
Solution Approach 1:
The patent performs preliminary action by training the surrogate model on a limited set of expensive model evaluations before actual predictions are needed. This upfront investment in creating the surrogate model enables rapid predictions later, significantly reducing the time required for subsequent estimations.
Solution Approach 2:
The surrogate model serves as a rapid copy that can be queried instantaneously compared to the time-consuming original model. This copying approach eliminates the need to repeatedly execute expensive experiments, dramatically reducing loss of time while maintaining prediction accuracy.
3Reliability
If expensive experiments are conducted to reduce uncertainty, then reliability is improved, but loss of substance increases due to large amounts of data required
Solution Approach 1:
The surrogate model creates a compressed representation (copy) of the input-output relationships that captures essential patterns with far fewer data points than traditional methods require. This copying approach enables reliable uncertainty quantification without needing large volumes of expensive experimental data.
Solution Approach 2:
The patent changes the parameter representation by using latent space embeddings and probabilistic formulations that efficiently capture uncertainty with reduced data requirements. By transforming the problem into a probabilistic framework with learned parameters, the system achieves reliable uncertainty quantification without requiring large datasets.
Data Source
AI summary
A system for intelligent selection of inputs for inference and quantification of uncertainty for active learning in a model includes computing an information gain by combining one or more metrics of interest and uncertainties in the inputs using information compression based on a first set of data as one or more pairs of values of inputs and outputs. One or more of the inputs are selected, with a potential information gain higher than other inputs, to query the model and record values of one or more of the outputs. The model computes a measurement of performance outputs based upon the first set of data. The model is updated based upon estimated metrics of interest performance and quantification of the associated uncertainty, and output for a change to a process based upon the estimated metrics of interest performance and quantification of the associated uncertainty.