Matrix Decomposition Search Optimization via Gradient-Based Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The MCMC method for searching solutions to matrix decomposition problems of binary matrices is time-consuming due to the flat shape of the evaluation function, making local searches difficult.
Innovation Solution
A data processing program that acquires and updates values of matrix elements in a matrix decomposition problem, determining changes in the evaluation function's value based on inequality constraint terms to efficiently search for solutions by altering matrix elements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the MCMC method is used to search for solutions to matrix decomposition problems, then the solution can be found, but the search takes a long time due to the flat shape of the evaluation function
Solution Approach 1:
The patent changes the parameter representation from binary matrix elements to real-valued parameters that can be continuously optimized. By transforming the evaluation function to have a non-flat shape with clear gradients, the optimization process can efficiently navigate the search space and find solutions much faster than traditional MCMC methods.
Solution Approach 2:
The patent replaces the stochastic MCMC search mechanism with a deterministic gradient-based optimization approach. This substitution eliminates the random walk through the solution space and replaces it with directed optimization along gradients, dramatically reducing search time while maintaining solution reliability.
2Ease of manufacture
If the evaluation function uses logical sum operations for matrix multiplication, then the matrix decomposition problem can be formulated, but the evaluation function becomes flat and local search becomes difficult
Solution Approach 1:
The patent transforms the evaluation function from a step-function-based logical sum to a continuous function with smooth gradients. By using squared differences or absolute differences instead of logical operations, the evaluation function becomes non-flat, enabling effective gradient-based optimization and local search while maintaining the same problem formulation capability.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A data processing program for causing a computer to execute: acquiring values of first matrix elements being problem information of a matrix decomposition problem representing a binary first matrix represented by the first matrix elements by a matrix product of a second and third matrix, and initial values of second matrix elements of the second matrix and third matrix elements of the third matrix; determining whether to adopt a change in a value of a fourth matrix element of any one of the second and third matrix elements based on a change amount of a value of an evaluation function; and searching for values of the second and third matrix elements by repeating processing of updating the value of the fourth matrix element while the fourth matrix element is changed when it is determined to adopt the change in the value of the fourth matrix element.