Optimum Solution Acquisition Using Variational Autoencoder Latent Space
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In complex solution spaces, existing optimization methods often result in increased captures and searches of local solutions, taking an enormous amount of time to reach an optimum solution, especially in multimodal spaces where the existence of optimization is unknown.
Innovation Solution
A non-transitory computer-readable storage medium storing a program that uses a machine learning model, specifically a variational autoencoder (VAE), to learn characteristic amounts of training data, calculate similarities, and acquire an optimum solution by concentrating data in a latent space, allowing for rapid acquisition of solutions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing optimization methods are used in complex solution spaces, then the search can cover the solution space, but the time required to reach an optimum solution becomes enormous
Solution Approach 1:
The patent applies preliminary action by pre-learning the landscape of the objective function using training data before actual optimization. The machine learning model is trained in advance to capture the characteristics and patterns of the solution space, so that during optimization, the search can leverage this pre-acquired knowledge to quickly identify promising regions without exhaustive exploration, thus resolving the contradiction between thorough search and time consumption
Solution Approach 2:
The patent introduces a machine learning model as an intermediary between the objective function and the optimization process. This intermediary learns the underlying patterns and characteristics of the objective function from training data, then guides the optimization search by providing informed predictions about solution quality. This mediator enables the system to navigate complex solution spaces efficiently without requiring exhaustive search, thereby reducing optimization time while maintaining solution accuracy
2Productivity
If the number of searches is reduced to rapidly acquire an optimum solution, then the time is reduced, but the ability to find the true optimum in multimodal spaces becomes uncertain
Solution Approach 1:
The patent performs preliminary learning of the objective function's landscape using training data before optimization. This pre-acquired knowledge about the solution space's characteristics, including multimodal features, enables the optimization process to make informed decisions with fewer searches while still reliably identifying true optima, thus resolving the contradiction between search reduction and optimization reliability
Solution Approach 2:
The patent implements feedback mechanisms where the machine learning model continuously learns from training data and adjusts its understanding of the objective function landscape. This feedback loop enables the model to improve its predictions about solution quality, allowing the optimization process to achieve reliable optimum finding with reduced searches by leveraging increasingly accurate guidance from the trained model
Data Source
AI summary
A non-transitory computer-readable storage medium storing a program that causes a computer to execute a process, the process includes obtaining a machine learning model having learned characteristic amounts of a plurality of training data including an objective function; calculating similarities between the characteristic amounts of the plurality of training data by inputting the plurality of training data to the obtained machine learning model; specifying a data group having a high similarity with a desired objective function from the characteristic amounts of the plurality of training data based on distances of the calculated similarities; and acquiring an optimum solution for the desired objective function by using the specified data group.