Neural Solver Initialization for Faster Multivariable Convergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In industrial environments, existing machine learning systems struggle to converge on solutions within the required time frame due to the complexity of multivariable problems, leading to missed opportunities for optimal scheduling and operation.
Innovation Solution
A system and method that utilize an appropriately trained machine learning software, such as a neural network, to generate a partial solution from input data, which is then provided to a mathematical solver to reduce the solution space and accelerate convergence to a full solution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning systems are used to solve multivariable problems in industrial environments, then solution accuracy can be improved, but the computational time required to converge on a solution increases
Solution Approach 1:
The neural network performs preliminary action by generating an initial solution estimate before the mathematical solver begins its optimization process. This pre-computed initialization point guides the solver toward the optimal solution more efficiently, reducing the iterative computation time while maintaining solution accuracy.
Solution Approach 2:
The neural network acts as an intermediary between the problem definition and the mathematical solver. It translates complex industrial process data into an initial solution estimate that the solver can refine, bridging the gap between raw data and optimized scheduling solutions.
2Loss of time
If the solution space is reduced to accelerate convergence, then computational time is reduced, but the quality and completeness of the solution may deteriorate
Solution Approach 1:
The neural network performs partial action by generating only an initial solution estimate rather than computing the complete optimized solution. This partial computation provides enough information to guide the solver, achieving rapid convergence without sacrificing solution quality in the subsequent refinement phase.
Data Source
AI summary
Significant gains in the execution speed of mathematical programming solvers may be made by applying a representation of the solver problem to a neural network and receiving from the neural network estimates of variables that can be used to initialize the solver. The time spent can be reduced by almost two orders of magnitude, allowing for faster turnaround of solutions, using less compute power, and increasing the ability to attack larger and more detailed problems. Graph neural networks can optionally be used effectively in this application. Subsets of solved problems may also be employed to train the neural network to make better estimates. This improved performance is of great value in producing more optimally efficient scheduling at a faster pace, as needed in semiconductor manufacturing scheduling.