Gilbreth supercomputer helping Purdue faculty make leaps in machine learning, artificial intelligence, data science

Purdue faculty working in machine learning, artificial intelligence and other fields that are optimized for computations run on graphics processing units (GPUs) have a powerful new resource in Gilbreth, Purdue’s newest community cluster research supercomputer.

Timothy Reese, a graduate student working with Michael Zhu, professor of statistics, is using Gilbreth to assist with his research in studying how deep neural networks perform the task of classification, a branch of machine learning that teaches computers how to sort objects into different groups – for example, looking at photos of dogs and determining which ones are English setters and which ones are Irish setters.

Reese’s work is especially well-suited for Gilbreth because he uses TensorFlow, deep learning software that is optimized for GPUs.

Because Reese couldn’t work nearly as efficiently on traditional CPUs – and setting up his own GPU cluster would be cost-prohibitive – the Gilbreth cluster has been key to his success.

“Gilbreth is a blessing,” says Reese. “Without it, I could not be doing this project.”

Gilbreth’s impact extends far beyond the 37 researchers who have purchased access to the cluster, however. It’s playing an important role in the success of Purdue’s Integrative Data Science Initiative (IDSI), says Sunil Prabhakar, professor of computer science and IDSI director.

The IDSI is a Purdue-wide initiative to help support and coordinate research, education and broader engagement in data science at Purdue, with a goal of bringing “data science to all” and ensuring that every Purdue graduate understands data science at a level that is appropriate for their discipline.

In addition to being used in some of the individual research projects that have been funded by the IDSI, Gilbreth plays a major role in the Data Science Consulting Service, a component of the IDSI that offers Purdue researchers hands-on support with data analysis and assistance extracting useful information from their data.

Guang Lin, an associate professor of mathematics and mechanical engineering who leads the Data Science Consulting Service, says that the projects his team typically tackles require a great deal of computational power, which is where Gilbreth comes in. 

In one collaboration, the team partnered with Wen Jiang, professor of biological sciences, to do three-dimensional reconstruction of the molecular structure of proteins and viruses from 2-D projections generated by cryo-electron microscopy.

“It’s an invaluable resource for researchers on campus that makes a big difference to what we are able to accomplish,” says Prabhakar of Gilbreth.

“The support that the Purdue community has received from the creation of clusters such as Gilbreth, but also the technical consulting support service that ITaP provides, has been an extremely useful resource.”

Access to Gilbreth is available through a subscription model that gives researchers access to a large pool of nodes for a low annual fee. Dedicated nodes may be purchased as well, as with previous community cluster systems.

To learn more about Gilbreth or the Community Cluster Program, contact Preston Smith, ITaP’s director of research services and support, psmith@purdue.edu or 49-49729.

Writer: Adrienne Miller, science and technology writer, Information Technology at Purdue (ITaP), 765-496-8204, mill2027@purdue.edu

Last updated: Sept. 8, 2019