Genetic Algorithm Fitness History Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Genetic algorithms used in data mining often face challenges such as premature convergence to local optima and the inability to accurately determine the fitness of individuals due to incomplete testing, leading to optimization for 'lucky' rather than truly fit individuals.
Innovation Solution
Implementing an experience-layered elitist pool structure where individuals are stratified by their testing experience levels, allowing only those with sufficient experience to compete and ensuring that fitness estimates are accurate before selection for the elitist pool, and using fitness trial history to weight and filter individuals for deployment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If genetic algorithms evaluate individuals based on limited testing data to speed up the process, then productivity improves, but measurement precision of fitness deteriorates
Solution Approach 1:
The patent applies preliminary action by requiring individuals to complete a predetermined number of fitness evaluations before they are eligible to compete for the elitist pool. This preliminary testing phase ensures that fitness estimates are based on sufficient data before individuals enter the competitive selection process, resolving the contradiction between evaluation speed and fitness estimate accuracy.
2Productivity
If individuals are selected for elitist pool based on early fitness estimates, then productivity improves, but reliability of selection deteriorates
Solution Approach 1:
The patent implements preliminary action by establishing a minimum evaluation threshold that individuals must meet before becoming eligible for the elitist pool. This preliminary filtering ensures that only individuals with sufficiently reliable fitness estimates based on adequate testing can compete, thereby improving selection reliability without significantly compromising productivity.
Solution Approach 2:
The patent applies beforehand cushioning by requiring a buffer of additional evaluations beyond initial testing before individuals can enter the elitist pool. This cushioning period allows fitness estimates to stabilize and become more reliable, preventing premature selection of individuals whose fitness was overestimated due to limited sampling.
3Measurement precision
If the algorithm runs longer to improve fitness estimation accuracy, then measurement precision improves, but loss of time increases
Solution Approach 1:
The patent applies segmentation by dividing the evaluation process into distinct phases: an initial evaluation phase with a predetermined number of assessments, followed by eligibility determination for elitist pool competition. This segmentation allows the system to achieve sufficient measurement precision within a reasonable time frame by stopping evaluations at appropriate thresholds rather than running indefinitely.
Solution Approach 2:
The patent implements partial action by performing a sufficient but not excessive number of evaluations - specifically a predetermined number that provides adequate fitness estimation accuracy without unnecessary additional computation. This partial evaluation approach balances measurement precision requirements with time constraints by doing just enough testing to make reliable selections.
Data Source
AI summary
Roughly described, a data mining system includes a candidate gene database in which each candidate individual has a respective fitness estimate. A gene pool processor tests individuals from the candidate gene pool on training data and updates the fitness estimate of the individuals. A gene harvesting module deploys selected individuals from the gene pool. The gene pool processor includes a competition module which selects individuals for discarding in dependence upon their updated fitness estimate. The system maintains a fitness training history for each of the candidate individuals, identifying the data samples on which the individual has been tested. The historical information can be used to assist in any one or more of the following: competition among the individuals, avoiding re-testing of an individual on the same data sample, removing duplicate test data before merging fitness evaluations, improving gene pool diversity, and selecting individuals for deployment.