Neural Network Data Augmentation via Feedback-Guided Seed Mutation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data augmentation techniques for neural networks are often manual, rely on human intuition, or are performed randomly, failing to effectively improve model performance due to limited and scarce training data.
Innovation Solution
A method that involves running a validation dataset through a neural network, analyzing correct and incorrect predictions, mutating seeds corresponding to correct predictions, and evaluating the increased neural network coverage for the mutated seeds, iteratively refining the data augmentation process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If manual or random data augmentation techniques are used, then data diversity is improved, but training effectiveness deteriorates due to lack of targeted improvements
Solution Approach 1:
The patent implements feedback by analyzing the neural network's predictions and using this information to guide the data augmentation process. The system identifies correct and incorrect predictions, then uses this feedback to determine which seeds should be mutated and how, creating targeted augmentations that specifically address the network's weaknesses rather than applying random transformations.
Solution Approach 2:
The system performs self-service by automatically analyzing its own performance and generating its own training data without external intervention. The neural network's prediction results directly inform the augmentation process, allowing the system to self-optimize its training dataset based on its own performance metrics.
2Reliability
If more training data is collected, then model performance is improved, but data availability deteriorates due to finite and scarce data
Solution Approach 1:
The patent applies copying by creating multiple variations of existing data seeds through mutation. Instead of collecting new data, the system copies and transforms existing seeds to generate augmented training data that maintains the characteristics of the original data while providing additional training examples.
Solution Approach 2:
The system changes parameters of existing data by mutating seeds in different dimensions. This allows the generation of diverse training examples from limited original data by varying parameters such as noise levels, transformation types, and mutation intensity, effectively expanding the training dataset without requiring additional data collection.
3Productivity
If automated mutation processes are implemented, then productivity is improved, but control over augmentation quality deteriorates
Solution Approach 1:
The patent implements dynamics by making the mutation process adaptive rather than static. The mutation strategy changes based on the neural network's performance feedback - the system dynamically adjusts which seeds to mutate, what mutation types to apply, and the intensity of mutations based on real-time analysis of prediction correctness.
Solution Approach 2:
The system performs preliminary action by pre-analyzing the neural network's predictions before generating augmentations. This allows the system to plan the mutation strategy in advance based on identified weaknesses, ensuring that the subsequent data generation is targeted and purposeful rather than random.
Data Source
AI summary
In accordance with an embodiment, a method for augmenting training data for a neural network includes: running a validation dataset through the neural network to provide a first output; analyzing the first output of the neural network to determine first correct predictions and first incorrect predictions using a classifier; mutating seeds of the validation dataset corresponding to the first correct predictions; running the mutated seeds through the neural network to provide a second output; analyzing the second output of the neural network to determine second correct predictions and second incorrect predictions using the classifier; determining whether there is an increase in neural network coverage for mutated seeds yielding the second correct predictions; and performing steps of mutating the seeds, running the mutated seeds through the neural network, and analyzing the second output of the neural network for the mutated seeds yielding the second correct predictions.