The future of machine learning looks bright as NVIDIA has found a way to teach their AI with little datasets.
This new approach to machine learning may help in providing new uses to this technology especially in the medical sector where they can gather diverse data to accelerate diagnoses of rare diseases.
What is a generative adversarial network (GAN)?
In order to understand what these AI are doing, we need to know what a generative adversarial network is. This is a type of machine learning model that involves two neural networks that compete with each other called a generator and discriminator.
These two networks work together to create a self-learning AI that can create a realistic output. Use cases of this technology are shown in converting black and white images into color, creating a realistic image from text and even deepfake videos.
Adaptive Discriminator Augmentation Is A Game Changer
NVIDIA researchers were able to improve their GAN through Adaptive Discriminator Augmentation (ADA). What it does is it only gets a few training images instead of feeding the AI hundreds and thousands.
The role of ADA is it will randomly distort these images until such a time that the AI will self-learn and synthesize these images.
This new GAN approach now requires 10 to 20 times less training images as NVIDIA claims.
As far as gaming is concerned, this might be useful for those who are into modding. A lot of games in Steam offer Steam Workshop support wherein players can create a spin-off of their favourite games through these mods.
NVIDIA is definitely going in the right direction with this breakthrough. If they also manage to find ways to optimize game graphics with using less computing power, then that could be another huge discovery if that happens.