In this Intel on AI podcast episode: Healthcare workloads, particularly in medical imaging, require more memory usage than other AI workloads because they often use higher resolution 3D images. Deep learning (DL) models developed from these data sets require both high accuracy and high confidence levels to be useful in clinical practice, but this is incredibly data and compute intensive. David Ojika, Research Scientist at the University of Florida, joins the Intel on AI podcast to talk about his research focused on the use of accelerators for machine learning (ML) as well as heterogeneous computing using Intel FPGAs, CPUs, and GPUs for inferencing. He describes a project that he led between Intel and Dell EMC which illustrated how 2nd Generation Intel Xeon Scalable processors with Intel-optimized TensorFlow on a DellEMC PowerEdge server was a very suitable configuration to address 3D models being deployed for medical imaging analytics. David talks about how, with more than 1 TB of system memory available, 2nd Gen Intel Xeon Scalable enable researchers to develop large DL models that can be several orders of magnitude larger than those available on existing DL accelerators. He expresses how this work between the University of Florida, Dell EMC and Intel better enable the use of AI-based medical imaging to help detect and diagnose cancer using MRI and other medical imaging systems and can ultimately help save lives.

To learn more, visit: intel.ly/memorybottleneck

Visit Intel AI Builders at: builders.intel.com/ai