Trace Dataset Augmentation Using Variational Autoencoders
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automatic classifiers face challenges with data imbalance and small training data dimensions, particularly in binary classification tasks, leading to poor performance.
Innovation Solution
A system and method for data augmentation using oversampling techniques and variational autoencoders to generate synthetic data, enhancing training datasets for classifiers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data augmentation is performed using traditional methods, then training data quantity increases, but data imbalance and poor classifier performance persist
Solution Approach 1:
The patent transforms data augmentation from a data quantity problem to a feature space problem by changing parameters from raw data duplication to latent space transformation. Variational autoencoders learn optimal transformation parameters in the latent space, generating diverse yet realistic synthetic samples that address both quantity and quality requirements for balanced classification performance
Solution Approach 2:
The patent introduces variational autoencoders as an intermediary between original data and synthetic training samples. This intermediary learns the underlying data distribution and generates realistic synthetic samples, acting as a bridge that transforms limited real data into abundant training data while preserving class balance and improving classifier performance
2Reliability
If more training data is collected, then classifier performance improves, but data collection time and resources increase
Solution Approach 1:
The patent creates synthetic copies of real data through variational autoencoders in the latent space. These synthetic copies preserve the statistical properties and class distributions of original data while providing unlimited training samples, eliminating the need for time-consuming data collection and enabling rapid model training and iteration
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
A system and method for classification. In some embodiments, the method includes forming a first training dataset and a second training dataset from a labeled input dataset; training a first classifier with the first training dataset; training a variational auto encoder with the second training dataset, the variational auto encoder comprising an encoder and a decoder; generating a third dataset, by feeding pseudorandom vectors into the decoder; labeling the third dataset, using the first classifier, to form a third training dataset; forming a fourth training dataset based on the third dataset; and training a second classifier with the fourth training dataset.