Neural Network Generation Using Predictive Classifier for Faster Convergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural networks generated by genetic algorithms lack transparency, making it difficult to determine their reliability and performance, leading to slow convergence on reliable models due to the time-consuming evaluation of candidate networks.
Innovation Solution
A system that uses a trained classifier to predict the reliability and performance of neural networks, allowing for the adaptation of the neuroevolutionary process by adjusting mutation parameters or discarding low-performing networks, thereby accelerating the convergence on high-performing models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional genetic algorithms are used to evaluate candidate neural networks by passing the entirety of a testing data set through each network, then the reliability of evaluation is improved, but the time consumption and productivity deteriorate
Solution Approach 1:
The patent applies preliminary action by training a classifier model in advance using supervised training data that describes characteristics of reliable and unreliable neural networks. This pre-trained classifier then rapidly evaluates candidate neural networks during the genetic algorithm, avoiding the need to pass entire testing data sets through each candidate during evaluation. The classifier provides quick preliminary assessments that guide the genetic algorithm while maintaining evaluation reliability.
2Adaptability or versatility
If the genetic algorithm continues evolving candidate neural networks without adaptation, then the thoroughness of search is improved, but the convergence time on reliable models deteriorates
Solution Approach 1:
The patent implements feedback by using the classifier's evaluation results to dynamically adapt the genetic algorithm's mutation parameter. When the classifier identifies that candidate neural networks have certain characteristics associated with unreliability, the system adjusts the mutation parameter to increase diversity and explore different regions of the solution space. This feedback loop enables the algorithm to converge faster on reliable models while maintaining adequate search coverage.
Solution Approach 2:
The patent applies dynamics by making the mutation parameter adaptive rather than static. The mutation parameter changes dynamically based on the classifier's assessment of candidate neural network characteristics. This dynamic adjustment allows the genetic algorithm to respond to the evolving population and guide the search more effectively toward reliable solutions, reducing convergence time while preserving search versatility.
3Speed
If the mutation parameter is adjusted based on classifier predictions, then the convergence speed is improved, but the complexity of the system deteriorates
Solution Approach 1:
The patent uses an intermediary approach by introducing a trained classifier model as a mediator between the genetic algorithm and the mutation parameter adjustment. The classifier, pre-trained on supervised data, serves as an intermediate layer that translates neural network characteristics into guidance for parameter adjustment. This intermediary simplifies the overall system complexity compared to implementing complex real-time analysis, as the classifier handles the analytical complexity during the training phase and provides straightforward predictions during execution.
Data Source
AI summary
A method includes determining, by a processor of a computing device, an expected performance or reliability of a first neural network of a first plurality of neural networks. The expected performance or reliability is determined based on a vector representing at least a portion of the first neural network, where the first neural network is generated based on an automated generative technique (e.g., a genetic algorithm) and where the first plurality of neural networks corresponds to a first epoch of the automated generative technique. The method also includes responsive to the expected performance or reliability of the first neural network failing to satisfy a threshold, adjusting a parameter of the automated generative technique. The method further includes, during a second epoch of the automated generative technique, generating a second plurality of neural networks based at least in part on the adjusted parameter.