RT Impact: Indiana University
Supporting artificial intelligence, machine learning, and GPU computing
Explosive impact
The Carbonate supercomputer was expanded to include 24 GPU nodes containing a total of 96 NVIDIA V100 GPUs, proving invaluable in both Artificial Intelligence research and more traditional HPC simulation research. For example, Professor Chuck Horowitz (IU Physics) and Assistant Professor Matt Caplan (Illinois State University) have a new theory of how supernovae explode, and they are testing it on Carbonate’s GPUs.
“Our simulations rely on the massive parallelism of the GPUs, as each nucleus must have its interactions calculated with every other. Without them, this work simply wouldn’t be possible.” – Matt Caplan, assistant professor,
Illinois State University
Mitigating Abuse with Natural Language Processing
Dr. Sandra Kuebler’s research team employs natural language processing for abusive language detection on social media. Finding abusive posts is increasingly important, but reliable datasets are crucial to creating machine learning approaches. Kuebler’s group investigates questions such as: Can we trust existing datasets to reflect the situation on social media? Do we introduce biases into the datasets by making decisions on how to sample the data? Does the language in abusive posts change over the years? Can we ask users to mark which posts on social media are abusive? Where are the boundaries between abusive and non-abusive posts? Do machine learning approaches pick up on biases from all of those decisions? How do we create unbiased datasets?
“Finding answers to all these questions involves developing and testing a range of models, most of them involving deep neural networks. Our work would not be possible without the high performance computing clusters, and more specifically the Carbonate GPU partition.” – Dr. Sandra Kuebler, Department of Linguistics,
College of Arts and Science, IU
In FY21 the Carbonate Deep Learning resource delivered 899,598 core hours and 97,723 GPU hours to users conducting research in disciplines including medical image segmentation, video classification, cybersecurity, genomics, and natural language processing.