Surrogate Model Optimization for Mixed Input Design
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional black-box optimization processes in materials science and semiconductor engineering are inefficient due to the need for numerous trial-and-error experiments, especially when dealing with mixed continuous-categorical inputs, and lack the ability to evaluate complex parameters effectively, requiring extensive resources and time.
Innovation Solution
A computer-implemented method using a rectified linear activation function (ReLU) deep neural network and mixed-integer linear programming to generate a surrogate model, allowing for the optimization of machine learning processes with mixed inputs, thereby reducing the number of experiments needed and improving prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional trial-and-error black-box optimization is used to search for optimal materials, then comprehensive exploration of the design space is achieved, but the process requires numerous experiments and excessive time and resources
Solution Approach 1:
The patent applies preliminary action by using Latin Hypercube Sampling to pre-generate a comprehensive initial dataset that covers the entire design space before optimization begins. This preliminary sampling ensures that the surrogate model is trained on representative data from all regions of the parameter space, enabling accurate predictions without requiring extensive trial-and-error experiments during the optimization phase.
Solution Approach 2:
The patent creates a surrogate model that copies the behavior of the complex black-box function. This surrogate model is trained on the initial dataset and can predict material properties quickly without requiring actual physical experiments or complex computations, thus reducing optimization time while maintaining prediction accuracy.
2Adaptability or versatility
If mixed continuous-categorical inputs are evaluated in conventional optimization, then comprehensive parameter evaluation is achieved, but the number of required experiments increases significantly
Solution Approach 1:
The patent applies universality by designing a unified optimization framework that can handle both continuous and categorical input types simultaneously. The Latin Hypercube Sampling method and surrogate model approach work universally for mixed input types, eliminating the need for separate evaluation processes for different parameter types and improving productivity.
Solution Approach 2:
The patent transforms categorical parameters into a format suitable for the surrogate model by encoding them appropriately. This parameter transformation allows the mixed-integer linear programming solver to efficiently evaluate both continuous and categorical variables together, reducing the number of required experiments while maintaining comprehensive parameter evaluation.
3Measurement precision
If extensive historical data is collected for training, then model accuracy is improved, but data processing time and computational resources increase
Solution Approach 1:
The patent extracts only the most essential features and parameters from historical data using Latin Hypercube Sampling. Instead of processing all available historical data, the method selectively samples representative data points that capture the essential variations in the design space, reducing computational resources while maintaining model accuracy.
Solution Approach 2:
The patent applies partial action by using a surjective mapping function that processes historical data in a computationally efficient manner. The surrogate model is trained on a carefully sampled subset of data that provides sufficient information for accurate predictions, avoiding the excessive computational cost of processing complete historical datasets.
Data Source
AI summary
A method and system of optimizing a machine learning process includes receiving an input set of historical data including input values and output values. The historical data is incorporated into a sampling design to form the initial dataset. A surrogate model of the machine learning model is generated by fitting the historical data using a rectified linear activation function (ReLU) deep neural network. Mixed-integer linear programming techniques are applied to the surrogate model to arrive at a set of predicted optimal inputs. The machine learning model is tested using the predicted optimal inputs. Output from the testing of the machine learning model is generated using the predicted optimal inputs. A determination from the output is made as to whether an optimal output has been generated by the testing of the machine learning model using the predicted optimal inputs.