GAN stands for Generative Adversarial Networks and is a framework for the development, training and refinement of Machine Learning models.
The Concept was or originally published in the paper arXiv:1406.2661v1 by a team of researchers at the University of Montreal in June 2014.
In this type of adversarial nets framework to models are pitted against one another in a min-max two player game. The two proposed adversary models are:
Generative Model: Think of a Counterfeiter trying to produce a fake (currency or artwork)
Discrminative Model: Think of the Detective trying to spot the fakes
The competition between the two models drive them to improve their respective methods till the counterfeits are indistinguishable from the originals.
In a recent article published by NVIDIA research GANs were used to create highly realistic scenes (photograph lookalikes) from simple block drawings (think a drawing in MS paint by a 5 year old)
You can read more about this awesome technology on the NVIDIA Research Blog
At Playware we can imagine such technologies being used for the authoring of photo-realistic environments for immersive simulated and virtual learning.
About the Author:
Siddharth Jain is the Creative Director for Playware Studios a Singapore Serious Games Developer. He develops games for Military, Healthcare, Airlines, Corporate and Government training and Mainstream education. He has taught game design in various college programs at NTU, SIM, NUS and IAL in Singapore and is the author and proponent of the Case Method 2.0 GBL pedagogy.
Comments open