Metamodel Efficiency Scoring for Detecting Suboptimal Process Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine-learning models are often locally optimized and can be globally suboptimal, making it difficult to detect using automated systems.
Innovation Solution
An apparatus and method for identifying collateral processes using a processor and memory to measure process data, generate a process model, and train a metamodel to output model efficiency scores, allowing for the detection of suboptimal models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If machine-learning models are locally optimized using training examples, then the model convergence speed is improved, but the model may become globally suboptimal and difficult to detect
Solution Approach 1:
The patent introduces a metamodel as an intermediary system that evaluates process models. The metamodel receives process model measurements and outputs efficiency scores, acting as a mediator between the locally optimized model and the evaluation criteria. This allows detection of globally suboptimal models while preserving the speed benefits of local optimization during training.
Solution Approach 2:
The patent implements a feedback mechanism where metamodel training examples include efficiency scores that relate to model input examples and output examples. This feedback loop allows the system to learn from efficiency evaluations and improve detection of globally suboptimal models over time, while maintaining the rapid convergence of local optimization during the training phase.
2Productivity
If automated systems are used to detect model optimality, then the detection efficiency is improved, but the ability to detect globally suboptimal models deteriorates
Solution Approach 1:
The metamodel serves as an intermediary evaluation layer that processes model measurements and generates efficiency scores. This intermediary system enables automated detection while improving the accuracy of detecting globally suboptimal models by providing a specialized evaluation mechanism rather than relying on standard automated detection methods.
Solution Approach 2:
The patent performs preliminary evaluation by generating metamodel training examples with efficiency scores before final model deployment. This preliminary action allows the system to pre-learning efficiency patterns and detect potential global suboptimality issues before they manifest in production, thereby improving both detection efficiency and accuracy.
3Measurement precision
If process models are trained using process input data and correlated output data, then the model training accuracy is improved, but the complexity of the overall system increases
Solution Approach 1:
The patent segments the evaluation system into distinct components: process models for data processing, metamodels for efficiency evaluation, and training example generators. This segmentation allows each component to specialize in specific tasks, maintaining training accuracy while managing system complexity through modular architecture where each segment handles a specific aspect of the overall process.
Data Source
AI summary
An apparatus and method identifies collateral processes. [T]he apparatus including a processor configured to measure a plurality of process data of a process, generate a first process model using the plurality of process data, receive a plurality of metamodel training examples, train, using the plurality of metamodel training examples, a metamodel, generate a measurement of the first process model and output a model efficiency score of the first process model using the metamodel and the measurement of the first process model.