Page 37 - GCN, August/September 2018

system, Schretzman said.
So where does machine learning
come into play? Those six groups were not defined before the project started. Instead, the team used the TraMineR package in the R statistical computing environment to do the analysis.
“Machine learning gives you the ability to direct the data to develop those clusters,” Schretzman said. “And the beauty [is] that the data itself is developing the clusters.”
Deep neural networks
The projects in New York City and Illinois use techniques that qualify as machine learning, but they’re at the simplistic end of the spectrum. The U.S. Bureau of Labor Statistics has also been using machine learning for years, and it is now preparing to make the leap into the use of deep neural networks.
Every year, BLS collects massive amounts of data. The Survey of Occupational Injuries and Illnesses, for example, includes 300,000 written descriptions of how workers have been injured. Those responses then have to be coded, which means, for example, making sure “reporter” and “journalist” are given the same code and that the injuries are placed in the correct category.
Until 2013, that coding was done manually and took about 20,000 hours to complete. “It’s not a fun thing to do,” said Alex Measure, a BLS economist who has been working on machine learning initiatives.
The agency started using machine learning by training a model on the historical hand-coded surveys. Now more than half of the coding is done by machine. Each night, completed surveys are run through an autocoding model, which also creates a probability for the accuracy of the coding result. If the probability is below a certain level, the survey is sent to an employee to code.
Such shallow machine learning has proven to be good at recognizing words or pairs of words, but it struggles with
strings of text. “Sometimes we have narratives where you really need to understand what a whole sequence of words mean,” Measure said.
In the phrase “no sign of concussion,” a shallow machine learning model might recognize the word “concussion” and even the phrase “sign of concussion.” But it might struggle to recognize that the “no” negates the “concussion,” he added.
Deep neural network techniques that can model complex non-linear relationships could help. BLS has been running its model on existing hardware using open-source software, such as Google’s TensorFlow library for dataflow programming. As the agency makes the transition into deep neural networks, however, it needs the processing power of an NVIDIA GPU server, Measure said.
Although that kind of computing power is available from cloud providers, he added that the sensitivity of the data requires BLS to manage its own hardware.
What the future holds
Machine learning might seem suitable foralmostanysituation,butthat’sfar from the case. The technology “is not a solution to every problem,” Measure said. “It’s a solution to problems where you have lots of training data and you don’t have an easier way to automate it.”
The technology is most easily applied to areas such as IT ticketing or call centers that receive a high volume of requests and therefore generate a significant amount of historical data, Delmolino said. Machine learning can make an impact on “anything where there’s high volume, long wait times or big backlogs,” he added.
Furthermore, Lipton warned that just because the technology can be doesn’t mean it should be used. He said certain applications, such as predictive policing, have the potential to perpetuate or even exacerbate existing biases thanks to the
model’s feedback loop.
If police are assigned to patrol areas
based on predictions “of where crimes are going to take place...you could wind up finding crimes where you look for them and, as a result, over-policing poor, disadvantaged neighborhoods, then finding more crime,” Lipton said. A model with biased sample data could “think a disproportionate number of crimes take place in those neighborhoods, and then it’s going to allocate even more police officers.”
Delmolino echoed the need to root out potential bias. Machine learning models require active management and tweaking to lessen bias over time.
“You can’t just buy a magic tool and deploy it,” he said. “You have to be aware of these things.”
One of the next big steps for machine learning could be the ability for multiple models to interact with one another and work together, Delmolino said. “I suspect we’re going to see some really fascinating requests for ‘How do I make sure my models work with each other? Is there a way for models to communicate with each other?’”
Another stepping-stone will be the integration of machine learning and robotic process automation. RPA can automate tasks such as transferring files, moving data from one field to another and other computer processes.
RPA is “pretty dumb, really, today,” said Craig Le Clair, a vice president and principal analyst at Forrester Research. But as machine learning becomes integrated with the technology, RPA will begin making more decisions independent of human involvement.
If machine learning is the brain, then RPA represents the limbs that provide the ability to reach different systems across an enterprise network and make changes as the brain sees fit.
“These are very general tools,” Lipton said of machine learning. “I think they find a large variety of use cases in any gigantic organization, and that includes government.”•
GCN AUGUST/SEPTEMBER 2018 • GCN.COM 37