acm-header
Sign In

Communications of the ACM

ACM TechNews

Opening the Black Box of Neural Networks


Water molecules.

Pacific Northwest National Laboratory researchers using deep learning neural networks to model water molecule interactions unearthed data about hydrogen bonds and structural patterns.

Credit: Ergorov Artem/Shutterstock

Pacific Northwest National Laboratory (PNNL) researchers used deep learning neural networks to model water molecule interactions, unearthing data about hydrogen bonds and structural patterns.

The PNNL team employed 500,000 water clusters from a database of more than 5 million water cluster minima to train a neural network, relying on graph theory to extract structural patterns of the molecules' aggregation.

The method provides additional analysis after the network has been trained, allowing comparison between measurements of the water cluster networks' structural traits and the predicted neural network, enhancing the network's understanding in subsequent analyses.

PNNL's Jenna Pope said, "If you were able to train a neural network, that neural network would be able to do computational chemistry on larger systems. And then you could make similar insights in computational chemistry about chemical structure or hydrogen bonding or the molecules’ response to temperature changes.”

From Pacific Northwest National Laboratory
View Full Article

 

Abstracts Copyright © 2020 SmithBucklin, Washington, DC, USA


 

No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account