The Patent
Machine learning methods and systems for identifying patterns in data using a plurality of learning machines wherein the learning machine that optimizes a performance function is selected.
Inventors:
Hemant Virkar (Potomac, MD, US); Karen Stark (Arlington, MA, US); Jacob Borgman (West Newbury, MA, US)
Class Name:
Data processing: artificial intelligence machine
Publication Date:
Feb 26, 2013
Publication #:
Claims
This patented technology when benchmarked against industry standard tools shows an enormous improvement in the speed at which end users can create predictive analytics. Digital Infuzion’s method took 11 seconds to conduct an analysis that ran all night using industry standard toolkits. By being able to almost instantaneously create predictive analytics, decision makers can directly benefit by extracting discoveries from their big data using the automated, efficient machine learning method that Digital Infuzion has invented.
How We’ve Used It
Digital Infuzion used this technology to train a learning machine on examples of gene expression for healthy muscle tissue and tissue samples showing sarcopenia. The optimal machine was able to provide over 94% success in predicting for other muscle samples prepped in a different lab, demonstrating a good ability to generalize new data and a robustness towards small changes in procedures and techniques in the lab. Further, the machine produced a list of highly ranked genes that differed from previously published work, which led to a new hypothesis about the disease. The results suggested that as muscle tissue ages, it loses some of its highly characteristic muscle specializations and becomes more similar to other types of tissue and undifferentiated cells. This is a novel concept for a muscle disease that typically occurs during old age and may suggest ground-breaking avenues of treatment.