Using AI to reshape the world

Passion // Project

It’s more than a job. These compulsively creative IBMers make innovation a way of life, creating amazing projects just for fun.

Share this page:

Using AI to reshape the world

Haytham Assem

IBM Master Inventor and Technical Manager for the Cognitive Computing Group, IBM Innovation Exchange
Dublin, Ireland

 

His passion project

Urban reasoning

Haytham Assem Haytham Assem

As principal investigator and technical manager of the IBM Innovation Exchange in Dublin, IBM Master Inventor Haytham Assem has a day job that touches on pioneering technologies like machine learning and ubiquitous computing. And when he’s not at work adding another patent to his resume (35 so far and counting), you are likely to find him pushing the technology envelope even further, using big data and deep machine learning techniques to see what insights can be gleaned from the social media habits of city dwellers.

Beyond Twitter and FourSquare

For the past few years, location-based social networks like FourSquare have enabled people to easily find restaurants, shopping centers, arts venues and other businesses in cities around the world. Planners identify functional regions such as business districts, residential neighborhoods, shopping areas and other components of a city as static functionalities that do not vary with time. But Assem felt these static functional regions didn’t go far enough to capture the shifting temperature of a vibrant city.

Discover what you can
do at IBM

Jobs at IBM

Using Manhattan as his laboratory, Assem developed a new way of identifying what he calls Temporal Functional Regions by using machine learning to mine huge volumes of social media data. “Every region in the city has a dynamic functionality through time, and for the first time, we can show this,” he explains. “In the morning it might be education, then in the evening it might be an area rich in restaurants and nightlife. This is a new level of granularity we’ve achieved in the big data era.”

“We are able to build cognitive models that show the way regions change in relation to time of day. The impact of this will be the ability of marketers to make highly personalized recommendations, based on these temporal functional regions.”

If you give me one person’s social media data, I would be able to predict, for example, that person is from Brooklyn.

“Also I believe that the new notion of temporal functional regions could help home buyers to understand, compare and contrast investment values for real estate, so an area or region with functionality that varies more than another indicates that it's home to a variety of services and activities, and its value is likely to be higher than a less dynamic area.”

A person walking in the morning through an area populated with schools might be targeted with advertising messages relevant to education or children’s activities. Someone else in that same neighborhood in the evening might see ads for local nightclubs and restaurants. The result benefits marketers, who can use precision targeting in their campaigns, and customers, who are more likely to see ads that speak to their interests.

Assem hopes urban planners will use his temporal function tagging to fine-tune their work as well. “Rather than saying, ‘okay, this is a business district, so I need to build an area nearby that has social services supporting that business district,’ they will be able to see in a much finer pattern, so the planning can be much more nuanced.”

Temporal functional regions

Social services
Eating
Recreation
Education
Traveling
Shoppping
Entertainment
Residence
Night life
Social services
Eating
Recreation
Education
Traveling
Shoppping
Entertainment
Residence
Night life

Following the crowd

Taking the process a little further, Assem has also developed a means of using social media activity to predict recurrent crowd mobility with various intensities in a large metropolitan area. Again using Manhattan as his focal point, he can discern regular rhythms of crowd movement through the city during a normal day.

From a marketing standpoint, these algorithms enable companies to target customers during times of high density. But other applications can literally be lifesaving.

“Being able to predict where the crowd normally shifts throughout the day is very useful for public safety, traffic management, disaster management and urban planning. If we are expecting a specific recurrent crowd mobility pattern for a particular day of the week and anything happens differently, it can raise an alarm calling for a quick investigation into why the pattern was interrupted. And being able to identify an outlier pattern like this quickly could mean that an ambulance doesn’t get stuck in an unexpected traffic jam.”

Toward the cities of tomorrow

Assem’s newest passion project uses deep learning to determine how residents of one metropolitan region differ from those in other regions in terms of social media behavior. He’s piloting this study using New York City’s five boroughs.

Working with data from Twitter, Foursquare and other popular applications, Assem can gauge when people first get on social media in the morning, what they’re doing at various times of day based on where they check in, when they stop tweeting at night, and so on.

Urban Reasoning

Toward the cities of tomorrow

Assem’s newest passion project uses deep learning to determine how residents of one metropolitan region differ from those in other regions in terms of social media behavior. He’s piloting this study using New York City’s five boroughs.

Working with data from Twitter, Foursquare and other popular applications, Assem can gauge when people first get on social media in the morning, what they’re doing at various times of day based on where they check in, when they stop tweeting at night, and so on.

We are able to build cognitive models that show the way regions change in relation to time of day.

“We use Insights for Twitter on Bluemix—the default stream is 10 percent of overall Twitter use,” he says. “The impact of this is that you will extract a different pattern in Brooklyn than you will in Manhattan. We may see that in Brooklyn, people tend to be up and on social media earlier. They may tweet about certain types of information at certain times, eat at a different time, go to bed at a different time than their counterparts in Manhattan.

Heat map

Heat map showing Manhattan
on a busy weekend

“If you give me one week of activities from all of the social media data we capture, we can turn that into a digital footprint. From that, if you give me one person’s social media data, I would be able to predict, for example, that person is from Brooklyn.”

The data can also identify areas of concern. If one neighborhood has a high level of people tweeting during working hours, or getting up later, that might indicate a higher level of unemployment in that neighborhood.

Looking more broadly, Assem believes these new findings could have implications for global development. “We have achieved 75 percent accuracy in New York City to extract unique patterns that differentiate between its five boroughs. If we apply a similar model to cities across the globe, we might find that cities in developing countries tend to follow one pattern, but maybe one of those cities has a pattern closer to that of London, and that raises a flag. It could indicate that this particular city is beginning to develop more first-world characteristics.”

“The system won’t tell us the root cause of the differences, but it will tell us that there’s something here to explore further. Maybe there are patterns related to economic factors, or to the education level of the population,” he muses. “The technology is still in the very early stages of development, but in my opinion, we’ll eventually be able to predict where cities will be in 10 years by continuously extracting these patterns and training the system to recognize them.”

“If you give me one week of activities from all of the social media data we capture, we can turn that into a digital footprint. From that, if you give me one person’s social media data, I would be able to predict, for example, that person is from Brooklyn.”

The data can also identify areas of concern. If one neighborhood has a high level of people tweeting during working hours, or getting up later, that might indicate a higher level of unemployment in that neighborhood.

Looking more broadly, Assem believes these new findings could have implications for global development. “We have achieved 75 percent accuracy in New York City to extract unique patterns that differentiate between its five boroughs. If we apply a similar model to cities across the globe, we might find that cities in developing countries tend to follow one pattern, but maybe one of those cities has a pattern closer to that of London, and that raises a flag. It could indicate that this particular city is beginning to develop more first-world characteristics.”

“The system won’t tell us the root cause of the differences, but it will tell us that there’s something here to explore further. Maybe there are patterns related to economic factors, or to the education level of the population,” he muses. “The technology is still in the very early stages of development, but in my opinion, we’ll eventually be able to predict where cities will be in 10 years by continuously extracting these patterns and training the system to recognize them.”

“For example, if we had this technology 10 years ago, we could have looked at Singapore. At that time, it wasn’t a very developed country, and so I expect that its pattern would have been very different from someplace like New York City. Eight years ago, we would have seen the patterns getting a little closer, then closer still at six years. At that point I’m able to say, ‘Ah, this country is going in the direction of city X.’”

“The power of this developing technology is the power of crowdsourcing. Each person is a sensor on his own, and from all those sensors, we can gather so much useful information. We refer to this as ‘human-as-a-sensor’.”

Moving toward urban reasoning

Assem is shifting his attention to an emerging generation of cognitive technologies that focus on reasoning rather than simple prediction. “Urban reasoning,” as he calls it, relies for input on other traditional fields like environmental engineering, civil engineering, network engineering, transportation and sociology in the context of urban spaces. Using that input in an AI model can help city managers and businesses fine-tune their plans.

The power of this developing technology is the power of crowdsourcing. Each person is a sensor on his own, and from all those sensors, we can gather so much useful information.

“The power of this developing technology is the power of crowdsourcing. Each person is a sensor on his own, and from all those sensors, we can gather so much useful information. We refer to this as ‘human-as-a-sensor’.”

Moving toward urban reasoning

Assem is shifting his attention to an emerging generation of cognitive technologies that focus on reasoning rather than simple prediction. “Urban reasoning,” as he calls it, relies for input on other traditional fields like environmental engineering, civil engineering, network engineering, transportation and sociology in the context of urban spaces. Using that input in an AI model can help city managers and businesses fine-tune their plans.

“Building new technologies for the aim of predicting and detecting events across cities is very interesting, but we can really have an impact once we’re able to understand the reasons for the patterns we detect,” he says.

A telco operator, for example, might be able to predict that a certain part of the city will be at its crowd peak between 9 a.m. and 4 p.m. on weekdays, but allocating a high level of resources to meet that crowd density might not make sense.

“If we know that the reason for the crowd pattern there is because it’s a business district, then we also know that most people in that area at that time are at work, where they have access to Wi-Fi, so crowd size doesn’t correlate to a high demand for network availability,” says Assem.

“On the other hand, if we know that the reason for an increased crowd size is because it’s an area where there are a lot of entertainment activities, then it makes sense to allocate more resources to meet the needs of that crowd.”

Inversely, planners could also use that kind of knowledge to spread those entertainment activities across a wider swath of the city, spreading out the stress on the city’s infrastructure as well.

“Urban computing aims to help us understand the nature of urban phenomena and even predict the future of cities,” Assem explains. “Urban reasoning aims to extend this vision with a main focus on providing insights about the reasons for the major challenges that our cities face, such as crowd congestion, increased network demand, air pollution and water resources management.”

For a deeper dive into Assem’s work, read “Spatio-Temporal Clustering Approach for Detecting Functional Regions in Cities,” published in Tools with Artificial Intelligence (ICTAI), 2016 IEEE 28th International Conference.

Share this page: