Argonne National Laboratory

Wozniak coauthors award-winning poster on machine learning and evolutionary computing

December 18, 2017

Justin Wozniak, a computer scientist in Argonne’s Mathematics and Computer Science Division, was coauthor of a poster that received a Best Poster Award at Mind Bytes, a research computing expo and symposium held in Chicago in May 2017. The award was presented in the area of scalability and performance.

Mind Bytes showcases work done using research computing resources at the University of Chicago and the Computation Institute. A major feature of Mind Bytes is the poster session, designed to facilitate the exchange of ideas across disciplines.

Wozniak and his coauthors won the award for their poster “High Performance Machine Learning and Evolutionary Computing to Develop Personalized Therapeutics.” In it, they presented their use of an evolutionary, or genetic algorithm (GA), to search large parameter spaces and determine intervention strategies for a patient’s dynamical immune system. Such algorithms are run until the starting population of interventions converges to a small set of near-optimal solutions. The work was funded by a NIH Big-Data to Knowledge (BD2K) grant (PIs: G. An, University of Chicago, C. Macal, GSS, Argonne).

For their study, the team used EMEWS (Extreme-scale Model Exploration with Swift), which combines existing machine learning/model exploration libraries with the Swift/T parallel scripting language to run scientific workflows in a high-performance computing environment (

“High-performance computing and EMEWS were essential in order to efficiently search the parameter space consisting of 1091 possible combinations of interventions” said Wozniak. “Our work utilized up to 2,000 processing cores simultaneously.”

The results showed that GA-derived intervention can be used to significantly reduce patient mortality rate. The researchers noted, however, that one limitation of genetic algorithms is that they use many fixed parameters – for example, a fixed length of interventions – and hence may be unable to adapt in real time to patients who do not respond to intervention. In such cases, a new genetic algorithm experiment can be run, but personalized intervention is needed in order to devise an alternative treatment. The researchers therefore suggest that future studies also consider other machine learning techniques such as deep reinforcement learning.

For further information, see Chase Cockrell, Jonathan Ozik, Nick Collier, Justin M. Wozniak, and Gary An, “High-Performance Machine Learning and Evolutionary Computing to Develop Personalized Therapeutics.”