Invertible Augmentation for GAN Training Stability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training generative adversarial neural networks (GANs) with limited data often leads to discriminator overfitting, causing training to diverge instead of converge, and requires large amounts of costly and challenging-to-collect data.
Innovation Solution
An adaptive discriminator augmentation mechanism is used to stabilize training in limited data regimes by applying an invertible augmentation operator to the discriminator's input distribution, preventing augmentation leakage into generated images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If large amounts of training data are used to train GANs, then the discriminator overfitting is reduced and training convergence is improved, but the data collection cost and complexity increase significantly
Solution Approach 1:
The patent creates synthetic copies of training data by applying invertible augmentation operators to the original data distribution. These synthetic augmented samples are generated during training and fed back to the discriminator, effectively copying and multiplying the limited original data to improve training convergence without requiring additional real data collection
Solution Approach 2:
The patent changes the parameters of data transformation by introducing invertible augmentation operators that modify the input data distribution. By adjusting these augmentation parameters dynamically during training, the system adapts the data characteristics to prevent discriminator overfitting while maintaining training convergence, replacing the need for large static datasets
2Quantity of substance
If standard dataset augmentation is applied to GANs, then additional training data is provided, but augmentation leakage occurs where the generator learns to generate images with the augmentations
Solution Approach 1:
The patent introduces an intermediary mechanism where invertible augmentation operators are applied only to the discriminator's input data, not to the generator's output. This intermediary layer allows the discriminator to see augmented versions of real data for training, while the generator receives clean data and learns to produce images without the augmentations, preventing leakage
Solution Approach 2:
The patent segments the data processing pipeline into separate paths: one path applies invertible augmentations to real training data for discriminator training, while another path keeps the generator's input data clean. This segmentation ensures that the discriminator benefits from augmented data diversity while the generator learns the true data distribution without contamination from augmentation artifacts
Data Source
AI summary
Embodiments of the present disclosure relate to a technique for training neural networks, such as a generative adversarial neural network (GAN), using a limited amount of data. Training GANs using too little example data typically leads to discriminator overfitting, causing training to diverge and produce poor results. An adaptive discriminator augmentation mechanism is used that significantly stabilizes training with limited data providing the ability to train high-quality GANs. An augmentation operator is applied to the distribution of inputs to a discriminator used to train a generator, representing a transformation that is invertible to ensure there is no leakage of the augmentations into the images generated by the generator. Reducing the amount of training data that is needed to achieve convergence has the potential to considerably help many applications and may the increase use of generative models in fields such as medicine.