paint-brush
Leveraging General Adversarial Networks for Material Sciencesby@jdbohrman
161 reads

Leveraging General Adversarial Networks for Material Sciences

by James D. BohrmanNovember 11th, 2020
Read on Terminal Reader
Read this story w/o Javascript
tldt arrow

Too Long; Didn't Read

Material scientists often face the challenge of figuring out how to effectively search the vast chemical design space to locate the materials with their desired properties. The use of GANs can be leveraged to generate hypothetical materials not existing in the training dataset, reaching a novelty of 92.53% when generating 2 million samples. The implications of general adversarial networks extend far beyond the applications in material science, but I have admit that this is one of the use cases I have really would have thought of applying AI to.

Company Mentioned

Mention Thumbnail
featured image - Leveraging General Adversarial Networks for Material Sciences
James D. Bohrman HackerNoon profile picture

Material scientists often face the challenge of figuring out how to effectively search the vast chemical design space to locate the materials with their desired properties. To address this challenge, many scientists have turned to artificial intelligence in the race to discover new and advanced materials.

Leveraging GANs for Efficient Sampling of Chemical Space

A general adversarial network is a variety of machine learning framework that leverages the idea of "adversarial training" where a network is trained on adversarial examples. It is an idea that originates from game theory and introduced to the machine learning community in 2014 by Ian J. Goodfellow. A GAN is comprised of three parts:

  •  A generator model for generating new data
  • discriminator model for classifying the generated data
  • The adversarial network that pits the two models against each other.


With this in mind, a targeted strategy for developing novel chemical compositions is to develop sampling algorithms that can exploit both explicit chemical knowledge and implicit rules of composition embodied in a large database of materials.

The proposed outcome would be a generative machine learning model based on a generative adversarial network (GAN) for efficient generation of new hypothetical inorganic materials. The use of GANs can be leveraged to generate hypothetical materials not existing in the training dataset, reaching a novelty of 92.53% when generating 2 million samples.

Conclusion

The implications of general adversarial networks extend far beyond the applications in material science, but I have to admit that this is one of the use cases I have really would have thought of applying AI to.

If you think this is cool, you should definitely take a look at the site that inspired me to write about this topic! It's a site featuring GAN generated chemicals that do not exist. Pretty neat. Thanks for reading!