Genetic Algorithm With Dynamic Fitness Threshold
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Prior genetic algorithms converge to local maxima due to static environments, failing to traverse global sample spaces effectively, particularly in predicting game outcomes where dynamic adjustments are needed to favor the user.
Innovation Solution
The ARON System modifies genetic algorithms with a dynamic fitness threshold and weighted training examples, integrating neural network architecture to evolve populations and predict game outcomes by transforming raw data into higher dimensional spaces, avoiding local maxima through self-analyzing metrics and optimal iteration settings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a static fitness metric is used in genetic algorithm, then the algorithm converges quickly to a solution, but it converges to local maxima rather than global solution
Solution Approach 1:
The patent implements a dynamic fitness threshold that evolves alongside the population, transforming the static fitness metric into a dynamic one. This allows the algorithm to adapt its assessment criteria during evolution, preventing premature convergence to local maxima while maintaining efficient traversal of the sample space. The dynamic threshold adjusts based on population performance, enabling continuous optimization toward global solutions.
2Use of energy by moving object
If gradient descent is used for neural network training, then the training process is computationally efficient, but it converges to local maxima and requires labeled training data
Solution Approach 1:
The patent replaces gradient descent (a mechanical optimization method) with a genetic algorithm that uses evolutionary operators. This substitution eliminates the need for labeled training data and avoids convergence to local maxima, as the genetic algorithm traverses the sample space through mutation, selection, and crossover operations rather than following gradient paths.
Solution Approach 2:
The system uses unsupervised learning where the neural network learns from unlabeled data through evolutionary optimization. The genetic algorithm automatically discovers optimal weight configurations without requiring external labels or supervision, making the system self-sufficient in terms of training data requirements.
3Productivity
If standard genetic algorithm operations are used, then the sample space is traversed efficiently, but the algorithm cannot adapt to dynamic environments or manipulate expected value in favor of the user
Solution Approach 1:
The patent introduces a dynamic fitness threshold that evolves with the population, enabling the algorithm to adapt to changing environments. This dynamic component allows the system to manipulate expected values and adjust to new conditions while maintaining the efficiency of standard genetic algorithm operations for traversing the sample space.
Data Source
AI summary
Embodiment method and associated apparatus relate to altering the expected value of a system modeled by a random process simplified to produce a binary outcome. Various embodiments modify a genetic algorithm to optimize such settings as population size, number of iterations to convergence, mutation chance, and sample space. Some embodiment ARON implementations correctly predict game outcome relative to the spread, based on transforming unrelated raw data, applying the transformed raw data to a modified genetic algorithm, generating multiple expected outcomes determined by the modified genetic algorithm as a function of the transformed raw data, and filtering the outcomes as a function of predefined metrics to produce a single end result that can be utilized effectively by an evolutionary-style algorithm. Various embodiment implementations use modified genetic algorithms with embedded neural network architecture to model and predict for a user a discrete forecast of a game-like scenario using selectively processed historical data.