From reactive to proactive
Apart from improving simulations, data collected from the Array of Things and other sources can also directly help planners in the day-to-day running of a city. For example, the City of Chicago used data analytics to decide how to prioritize food safety inspections.
“The city has 35 inspectors who have to inspect 15,000 stores and restaurants ideally every year, which works out to 500 inspections per year, per inspector,” Catlett said. “Chicago’s data scientists worked with several companies to build a machine learning model, and we were approached to help make this model more efficient and effective.”
The model ranks each food establishment as high, medium or low in terms of their risk of food safety violations. This ranking is based on over 30 leading indicators, including the results of previous inspections, the size of the restaurant or the neighborhood the restaurant is in. Instead of selecting establishments to inspect more or less by random, the inspectors now order the inspections based on risk, which has resulted insignificant improvement in how quickly safety problems are uncovered and addressed.
“Another indicator that might be counterintuitive until you talk to people familiar with it, is whether the restaurant has a liquor license or not,” Catlett shared. “If it has a liquor license, that lowers the risk because they know that if they have a food safety violation, the city will threaten to take their liquor licence, which is a very valuable thing.”
In a 60-day double blind test, the team found that the risk-based inspections were finding food safety violations about a week sooner than they would have otherwise. The model has now become part of the city’s operational inspection process, generating a list of places to inspect every morning.
“This approach applies not only to food safety but any kind of inspection, such as elevator inspections or fire safety inspections,” Catlett said. “When our team got involved, we introduced a framework called AutoMOMML, developed by Argonne computer scientists, that allows people who are not data scientists to evaluate their variables and see which ones are co-dependent so that their models can be simplified, and tests the performance of multiple machine learning techniques. With this framework for testing models, we estimate that we can save the data analytics team several months of work on any given problem.”
“What we’re trying to do is figure out how we can use data science to get out in front of problems, which was a goal articulated by Chicago’s previous mayor, Rahm Emanuel. The goal is to move from reactive policies and measures to spotting trends and taking proactive measures,” he said.
Democratizing data
For Catlett, the real potential of HPC in smart cities is making the city a more livable place for all. Citing the example of the ‘rat patrol,’ which responds to non-emergency calls reported through the 311 line, Catlett said that the frequency of calls is typically a poor indication of where the problems are.
“For example, in a city like Chicago, there are potholes every spring after the roads thaw out. If you just went by 311 calls, you would think that the more wealthy parts of the city had worse roads, as they are the ones calling about potholes,” he said. “In reality, the poorer parts of the city actually have worse roads; it’s just that they don’t call 311 about them.”
When the city adopted a model for predicting rat infestations rather than depending solely on calls, they came across the worst rat infestation in recent history in a part of the city that doesn’t normally call 311.
“That was a real turn of the corner with big data because it showed that the new processes could be more equitable than the traditional processes and the city is now able to go to places where the problems are. “To me, that was an important contribution that big data has made to cities,” Catlett said.
Nonetheless, Catlett acknowledges that data can be dangerous if used improperly. This is why he says every research project should begin with two questions: what is it that you’re trying to accomplish, and is it an ethical thing to do in the first place?
“Beyond accomplishing your goal, you also need to think of ways that the capability could potentially be misused,” he said.
Though most of the measurements from Array of Things are about the physical environment or aggregate flow of people and vehicles, the policies and mechanisms that would protect privacy with respect to images was one of the key issues raised during public consultations, underlining how privacy is a key concern for citizens where smart cities technologies are being deployed.
To ensure the safety of images and sound data collectedby the sensor nodes, the team adopted HPC on the devices themselves, analyzing the images to extract information such as the number of people or vehicles and then deleting the images rather than transmitting them to be analyzed by centralized servers and saved.
“This is HPC at the edge. By doing all our artificial intelligence at the edge, we can have this very strong, auditable and verifiable privacy policy,” Catlett explained. “We also wanted to make decisions at the edge with a latency that was short enough for smart city research where infrastructure might be controlled in real time. Here, if you send information back to a server and wait for an answer it will take too long. This was particularly important to transportation researchers who wanted to experiment with intelligent intersections and vehicle-to-infrastructure communication.”
Whether out at the edge or back in a supercomputing center doing more simulations, HPC is going to be critical in almost every aspect of making cities smarter.
“It’s an exciting time to be in HPC and we’re just getting started,” he concluded.
This article was first published in the print version of Supercomputing Asia, January 2020.
Click here to subscribe to Asian Scientist Magazine in print.
———
Copyright: Asian Scientist Magazine; Illustration: Lam Oi Keat/Supercomputing Asia.
Disclaimer: This article does not necessarily reflect the views of AsianScientist or its staff.
