MAKING MACHINE LEARNING ACCESSIBLE TO ALL
“Many call this the age of information” says Rajive Ganguli, the Malcolm McKinnon Professor of Mining Engineering at the University of Utah. “It is perhaps more accurate to call it the age of data since not everyone has the ability to truly gain from all the data they collect. Many are either lost in the data or misled by it. Yet, the promise of being informed by data remains.”
Ganguli who is also the College of Mines and Earth Science’s associate dean (assessment) is launching UteAnalytics, a new, free analytics software which makes artificial intelligence (AI) or machine learning (ML) accessible to all.
Founder of the ai.sys group at the U, Ganguli says that as long as a client knows their data (that is, is an expert in their domain), they can use UteAnalytics to understand better the problems they are trying to solve. The research group’s mission is to seek insight from data, models systems and to develop computational tools for education and research.
At various points in time, Ganguli has developed ML tools that his students could use in class. Years ago it occurred to him that more could benefit from ML if only his workflow and tools were more user-friendly. His vision was finally brought to fruition by graduate student Lewis Oduro MS’23 who leveraged the numerous public domain ML tools available to programmers and converted the concept into Windows-based software.
“The tool is problem agnostic,” Ganguli says. “Hence it can have a broad group of users. I have used it for a variety of projects I am involved in, including mining, atmospheric sciences/air quality and COVID/hospital admissions.”
He reports that tens of subject matter experts (SMEs) who are non-coders have already subscribed to receive the software in advance of its formal release. “Many are professionals across a broad spectrum of fields from social science to business,” along with scientists and engineers.
Master of a domain
Potential clients for UteAnalytics may be “master of their domain,” with large sums of data in various formats and curious about what insights they can gain with ML, but they know nothing about it and certainly not about how to code ML or even do basic data analysis.
Designed to empower the domain expert, UteAnalytics allows a client to clean their data (remove nulls, convert data type from string to numeric, etc.); apply filters, to consider data within specific magnitudes, among other functions; conduct exploratory data analysis on the data; and apply linear regression, random forests (regression and classification) and neural networks (regression and classification).
The software also allows users to estimate effect of each feature (input) on the output as well as develop models in advance of predicting on a new dataset.
Daniel Mendoza, who holds faculty appointments in the Department of Atmospheric Sciences and elsewhere at the U, is an early adopter of the software. Through his work with air quality monitors on UTA trains and electric buses in Salt Lake Valley he and his team have successfully collected data for over 8 years for particulate matter and ozone data and now nitrogen oxides.
“When we look at neighborhood-specific data we can drill in and really see some social justice impacts,” Mendoza reported last year. Today, he is “using UteAnalytics to quickly and efficiently analyze the temperature data that we’ll be collecting in real-time from our mobile and stationary sensors. UA ,” he says, “gives researchers the power to look at data in a very streamlined way without endless hours of coding. The included tools facilitate a thorough interpretation of data and save time without compromising reliability.”
The difference that data — assisted by UteAnalytics tools — make in Mendoza’s work on air quality is most recently seen in the Urban Heat Watch campaign, involving citizen scientists who are helping collect data along the streets of Salt Lake Valley. As one of the top three urban heat islands in the nation, the Salt Lake City metropolitan area features a groundbreaking monitoring program. In fact, no where else in the world is there an initiative that exists at this density and scale than in Utah’s capital city and environs. And now UteAnalytics is helping Utah’s clean air initiative as well.
An Auspicious Launch
UteAnalytics is just the latest deliverable for Ganguli who has led approximately $13M in projects as primary investigator. He is currently involved in several projects in five different countries — U.S., Denmark/Greenland, Mongolia, Saudi Arabia and Mexico — on topics ranging from ML to training. With the launch of UteAnalytics, it’s the fruition of a long-term ambition that, now available to the public, has an auspicious future ahead of it.
Meanwhile, Oduro, who defended his thesis this past spring, has since taken a job near Phoenix, Arizona as a mining engineer at Freeport-McMoRan, a leading international mining company. A native of Ghana, Oduro says of his mentor, “He gave me the chance to work under him and provided me with the kind of relationship only evident between a father and a son.” Under Ganguli’s tutelage and support, Oduro was the principal player in building UteAnalytics as desktop software used for data analytics and building predictive ML models.
“I will forever be indebted to him and to the entire faculty at the University of Utah’s Mining Engineering Department,” the young scientist says on his LinkedIN page.
By David Pace
SME’s who are curious about applying ML to their data sets
can now download UteAnalytics from the website.