/tag/machine-learning
Machine Learning and UVA HPC
Overview Many machine learning packages can utilize general purpose graphics processing units (GPGPUs). If supported by the respective machine learning framework or application, code execution can be manyfold, often orders of magnitude, faster on GPU nodes compared to nodes without GPU devices.
The HPC system has several nodes that are equipped with GPU devices. These nodes are available in the GPU partition. Access to a GPU node and its GPU device(s) requires specific Slurm directives or command line options as described in the Jobs using a GPU Node section.
Applications Several machine learning software packages are installed on the UVA HPC system.
TensorFlow and UVA HPC
Overview TensorFlow is an open source software library for high performance numerical computation. It has become a very popular tool for machine learning and in particular for the creation of deep neural networks. The latest TensorFlow versions are now provided as prebuilt Apptainer containers on the HPC system. The basic concept of running Apptainer containers on the HPC system is described here.
TensorFlow code is provided in two flavors, either with or without support of general purpose graphics processing units (GPUs). All TensorFlow container images provided on the HPC system require access to a GPU node. Access to GPU nodes is detailed in the sections below.
PyTorch and UVA HPC
Description PyTorch is a deep learning framework that puts Python first. It provides Tensors and Dynamic neural networks in Python with strong GPU acceleration.
Software Category: data
For detailed information, visit the PyTorch
website.
Available Versions The current installation of PyTorch
incorporates the most popular packages. To find the available versions and learn how to load them, run:
module spider pytorch The output of the command shows the available PyTorch
module versions.
For detailed information about a particular PyTorch
module, including how to load the module, run the module spider command with the module’s full version label. For example:
module spider pytorch/1.
Center for Diabetes Technology PriMed
In their research around constant glucose monitoring and the automated maintenance of insulin for patients, the CDT is exploring data drawn from external data sources such as DexCom and FitBit. RC has assisted the CDT by designing a secure computing footprint in Amazon Web Services to pull in these data, parse and process them, in order to perform deeper analytics through machine learning. In January 2018, CDT sponsored a ski camp at Wintergreen Resort for a group of youth diagnosed with Type I diabetes with the goal of importing glucose, insulin, and exercise metrics at the end of each day through remote web APIs.
Predicting ER Triage Levels with Machine Learning
Before patients are admitted to the emergency room, they are assigned a triage level based on the severity of their health problems. This is accomplished using the Emergency Severity Index (ESI), an emergency department triage algorithm that classifies patient cases into five different levels of urgency. Researchers are interested in using machine learning to develop a model to predict patient triage level. This model would not only analyze the typical vital signs that are used in the ESI, but also demographic data and patients’ history of health.
Demographic and health data have been collected. RC is helping to prepare and normalize the data for use in a machine learning model.
