Neural Reparameterization for Physical Design Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Optimizing physical design solutions within a design space is challenging due to limited automation in parameterization, requiring numerous trial and error iterations, and relying heavily on manual approaches, which restricts the application of advanced optimization techniques.
Innovation Solution
A computer-implemented method using a machine-learned model to reparameterize the design space, iteratively updating parameters based on a gradient of an objective function, allowing for the generation and optimization of proposed solutions within the design space, thereby enabling more efficient and automated optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If manual trial and error approaches are used to identify optimal design solutions, then flexibility in exploring design options is maintained, but the time required and number of iterations increase significantly
Solution Approach 1:
The patent introduces a machine-learned model as an intermediary between the design space parameterization and the optimization process. This model learns the mapping from design parameters to performance outcomes, enabling automated optimization while preserving design exploration flexibility. The model acts as a mediator that captures complex relationships without requiring manual trial-and-error iterations.
2Device complexity
If the design space is parameterized with a limited set of parameters, then the problem complexity is reduced, but the opportunity to apply automated optimization techniques is restricted
Solution Approach 1:
The patent transforms the design space parameterization by introducing a machine-learned model that learns optimal parameter transformations. Instead of using fixed, limited parameters, the system learns parameter mappings that enable more effective automated optimization while adapting to the specific design problem complexity.
3Reliability
If numerous trial and error iterations are performed to find optimal solutions, then thorough design exploration is achieved, but computational resources and time are consumed
Solution Approach 1:
The patent applies preliminary action by training a machine-learned model beforehand to capture the design space characteristics and performance relationships. This pre-learning phase enables subsequent optimization iterations to be more efficient, as the model provides informed guidance rather than requiring exhaustive trial-and-error exploration for each design iteration.
Data Source
AI summary
The present disclosure is directed to a system for reparameterizing of a neural network to optimize structural designs. The system can obtain data descriptive of a design space for a physical design problem. The design space is parameterized by a first set of parameters. The system can reparameterize the design space with a machine-learned model that comprises a second set of parameters. For a plurality of iterations, the system can provide an input to the machine-learned model to produce a proposed solution. The system can apply one or more design constraints to the solution to create a constrained solution. The system can generate a physical outcome associated with the constrained solution using a physical model. The system can evaluate the physical outcome using an objective function and update at least one of the second set of parameters. After the plurality of iterations, the system can output a solution.