Data Augmentation Using Linear Interpolation and Random Numbers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning systems face challenges in training models with a large number of parameters due to the limited number of input and output samples, leading to suboptimal performance, as current data augmentation methods do not effectively generalize the training data to cover diverse scenarios.
Innovation Solution
A data augmentation method that selects multiple sets of samples, generates random numbers, and combines input and output samples using linear interpolation, along with image processing techniques such as inversion, translation, and brightness adjustment, to generate new extended samples, effectively increasing the dataset size and diversity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If traditional data augmentation methods are used, then the dataset size is increased, but the data diversity and generalization capability are insufficient
Solution Approach 1:
The patent combines multiple existing data augmentation techniques (image processing transformations, noise addition, data mixing) into a unified multi-stage augmentation pipeline. This merging approach allows simultaneous application of various transformation types to achieve both quantity increase and diversity improvement that single methods cannot achieve alone
Solution Approach 2:
The patent creates composite augmented data by combining real samples with synthesized samples, and further combining multiple transformed versions of the same data. This composite approach produces training data with enhanced diversity and representativeness, resolving the contradiction between quantity and quality
2Reliability
If more training samples are collected, then the model performance is improved, but the cost and time consumption increase
Solution Approach 1:
The patent creates synthetic copies of existing training samples through various transformation operations (rotation, flipping, cropping, color jittering). These copied and transformed samples serve as effective training data without requiring additional data collection, thus improving model performance while avoiding time loss
Solution Approach 2:
The patent performs comprehensive data augmentation in advance before model training, creating a large pool of diverse training samples beforehand. This preliminary action eliminates the need for time-consuming data collection during model development and deployment phases
3Adaptability or versatility
If data augmentation is performed extensively, then the dataset diversity is improved, but the computational complexity increases
Solution Approach 1:
The patent divides the data augmentation process into multiple independent stages (geometric transformations, photometric transformations, data mixing, noise addition). Each stage can be applied selectively and independently, allowing computational resources to be allocated efficiently while achieving comprehensive data diversity
Solution Approach 2:
The patent implements selective augmentation where different augmentation operations are applied with different probabilities or intensities based on data characteristics and computational constraints. This partial action approach maintains data diversity while controlling computational complexity by not applying all possible transformations uniformly
Data Source
AI summary
A data augmentation method includes: selecting at least two different sets of samples from an original data set, each set of samples including input samples and output samples; generating at least one random number; generating at least one extended input data sample according to input samples in the at least two different sets of samples and the at least one random number; and generating at least one extended output data sample according to output samples in the at least two different sets of samples and the at least one random number, each extended input data sample corresponding to a respective extended output data sample.