Genetic Model Data Augmentation for DNN Training Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional deep neural networks (DNNs) face challenges in achieving high accuracy for classification tasks due to limited data augmentation with random variations, leading to inefficient training times and suboptimal performance across all possible data point variations.
Innovation Solution
The proposed solution involves using a genetic model to generate a population of data points from a seed data point, determining a best-fit data point through a fitness function, and iteratively reselecting this data point for subsequent epochs, thereby enhancing data augmentation and training efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If training is executed with all possible variations of data points, then classification accuracy is improved, but training time increases substantially
Solution Approach 1:
The patent applies partial action by generating only the necessary number of data variations required for effective training, rather than exhaustively creating all possible variations. The genetic model generates a controlled population of transformed data points, selecting optimal variations through fitness evaluation, thus achieving good classification accuracy without the prohibitive training time cost of exhaustive variation generation.
Solution Approach 2:
The patent changes the parameter of data variation generation from random conventional methods to genetically optimized transformations. By applying genetic operations (mutation, crossover, selection) to generate and evaluate data point variations, the system identifies the most effective transformations that maximize classification accuracy while minimizing the number of variations needed, thereby reducing training time.
2Measurement precision
If many variations of data points are applied for data augmentation, then classification accuracy across all variations is improved, but the size of training dataset increases substantially
Solution Approach 1:
The system generates only the necessary number of high-quality variations through genetic operations, rather than creating a large volume of random variations. The fitness function evaluates each generated variation to ensure quality, allowing the system to maintain a compact training dataset size while achieving comprehensive classification accuracy across different data variations.
Solution Approach 2:
The patent transforms the approach from random data augmentation to genetically optimized data generation. By applying genetic algorithms, the system produces a smaller number of highly relevant and diverse data variations that are optimally suited for training, thus improving classification accuracy without substantially increasing the training dataset size.
3Productivity
If limited number of random variations are used for training, then training time is reduced, but classification accuracy for all possible variations deteriorates
Solution Approach 1:
The patent changes the nature of data variations from random to genetically optimized. By applying genetic operations (selection, mutation, crossover) to generate data variations, the system ensures that even with a limited number of variations, each one is highly relevant and informative for the classification task, thereby maintaining high classification accuracy while preserving training efficiency.
Solution Approach 2:
The genetic model autonomously generates and evaluates data variations through fitness functions, automatically identifying the most useful transformations without requiring extensive manual intervention or large dataset sizes. This self-optimizing process ensures high classification accuracy with limited variations, maintaining both training efficiency and model performance.
Data Source
AI summary
According to an aspect of an embodiment, operations may include selecting, from a training dataset, a first data point as a seed data point. The operations may further include generating a population of data points by application of a genetic model on the seed data point. The population of data points may include the seed data point and a plurality of transformed data points of the seed data point. The operations may further include determining a best-fit data point in the generated population of data points based on application of a fitness function on the generated population of data points. The operations may further include executing a training operation on the DNN based on the determined best-fit data point. The operations may further include obtaining a trained DNN for the first data point based on the training operation on the DNN based on the determined best-fit data point.