Skip to main content
Article | Mathematics and Computer Science Division

MCS researchers participate in AI for Science” workshop

Several researchers in Argonne’s Mathematics and Computer Science (MCS) division participated in an AI for Science Workshop held at Argonne on August 2223, 2019.

The workshop was part of the newly launched AI for Science” initiative at Argonne aimed at accelerating the development and adoption of artificial intelligence approaches in scientific and engineering domains to enable breakthroughs in energy, basic science, medicine and national security.

On the first day of the workshop, MCS division computer scientist Prasanna Balaprakash discussed Deep Learning Basics.” He described the fundamentals of deep neural networks starting from perceptron; discussed key algorithmic approaches of training deep neural networks; and highlighted the methods to overcome underfitting and overfitting in deep neural networks.

Also on the first day of the workshop Nicola Ferrier, a senior computer scientist in the MCS division, presented a talk on Machine Learning at the Edge.” 

Enormous amounts of data are needed to develop and train a machine learning, or ML, model,” said Ferrier. Researchers are now exploring a strategy in which these tasks are done in the cloud and the results moved to specialized devices at the edge, where AI or ML functions can be performed rapidly and efficiently on new data.”

Day 2 of the workshop provided a closer look at the work of Argonne researchers in AI and deep learning. Balaprakash co-presented Recurrent Networks for Time Series Data.” Time series data forecasting is challenging because it often involves temporal dependencies. Balaprakash discussed the strengths and weaknesses of recurrent neural networks, originally developed for natural language processing, and how they can be used to capture long-term temporal dependencies in time series.

In addition to the presentations, Tanwi Mallick, a postdoctoral appointee in the MCS division, assisted in hands-on tutorial sessions. Mallick’s work include the development of scalable data-efficient ML methods for network congestion modeling, high-performance computing and traffic modeling.