In this Cognitive Business interview, we explore security artificial intelligence (AI), with Vergence’s CEO, Jon Vrabel, and CTO, Jason Wallace at Evonexus.
Jon Vrabel previously worked at analytics software company Palantir, whose clients included US federal security and intelligence agencies. Jason Wallace previously worked in depth with technology solutions for the nation's intelligence community. In 2014, Jon and Jason met, noticed their complementary skillsets and passions, and decided to co-found Vergence, a video AI platform for law enforcement and the department of defense.
I had the opportunity to catch up with them recently, and let me tell you: they are passionate about keeping you safe using AI. I invite you to learn more about the world of public safety AI technology.
What problem does your company solve?
The short answer is: security agencies today have challenges in leveraging their video feed to prevent, detect, and address security threats. This is all due to the limitations of searching through the unstructured data. Vergence helps address these issues.
Why is unstructured data a challenge?
Well, you see, most video has metadata that is structured for searching. Some examples are, video source, location, date and time. That’s easy to search for. The video content itself however is unstructured, making traditional ways of searching for an object or action within the video very difficult and extremely time consuming.
“We started Vergence because we believe that the public safety officers who put their lives on the line every day to to protect our families should have access to the best technologies available.” - Jon Vrabel
So, what exactly does Vergence do to help with security video?
We create the technology that helps leverage public safety’s video. We aggregate disparate video sources into a single interface, making the video easy to search. But we don’t stop there. Vergence also provides investigation collaboration tools. Essentially, we’ve created a platform where public safety organizations can ingest video data from both open source and secure streams and utilize our open framework for video analytics to deploy computer vision and machine learning applications at scale across all these video sources. This open framework liberates public safety’s video data and enables them to build their own deep learning models or work with researchers on new cutting edge solutions. This puts the power of highly available deep learning research in the hands of the customer as a plug and play option.
You can say Vergence helps public safety organizations by augmenting their ability to leverage their video assets, in a timely manner, to prevent, detect, and address security threats.
And what’s the technology behind Vergence?
We built Vergence to leverage an Amazon cloud infrastructure, providing multiple benefits for our customers who use traditional video management services (VMS) today. There are no onsite resources and hardware for them to manage that will inevitably be outdated, as well cloud storage costs today are lower than hosting your own storage.
There is still some discomfort with the security of cloud solutions. What are your thoughts on that?
Counter-intuitively, a cloud infrastructure is more secure than traditional hosted environments. With a cloud solution you know that the latest security patches and best practices are being deployed as they become available, whereas it is easy to miss security fix in a local environment. Scalable cloud resources also enable the possibility of incorporating massive machine learning solutions, centralized for collaboration across multiple agencies and researchers.
Let’s look under the hood. How does your solution work, from a cognitive computing architectural perspective (in laymen terms)?
VergenceNet: our proprietary neural network is focused on teaching machines to watch surveillance video. For example, an analyst tags objects in the Vergence platform that are of interest to them and the neural net ‘learns’ from these tags. Once enough tag data has been collected, the neural net can begin to automatically recognize these objects, reducing the analyst’s workload.
Switching gears, how do you define cognitive computing?
Cognitive computing is tossed around (along with AI) as a sexy way to talk about applications of algorithms that ‘learn’ over time. To us it is using deep learning libraries that model the way connections inside the brain process visual stimuli.
Computer models essentially learn to “see” by detecting the edges of an object, determining the boundaries of that object, and then classifying what that object is based on recognizing the patterns of those shapes. In our case, learning to identify interesting things within video that can help investigations.
I hear you. What trends and challenges are most impacting AI / cognitive computing, in public safety, right now?
The Advancement in GPU and specialty hardware are continually bringing costs down and making cognitive computing economical for large scale video processing. The software models to run cognitive tasks have been around for decades, it’s the recent leaps in hardware capability that have really enabled this field to take off. With modern GPU technology it has become feasible to run neural networks with the scale and speed to produce actionable results.
We see a public discussion occurring around how these types of technologies will be used in public safety. Any advancement in technology brings the opportunity for abuse, and just as in the past, we need to come together as a community and decide how it will be used and what safeguards need to be put in place.
What are your predictions about the future of AI / cognitive computing 2016 and beyond, in public safety?
Cognitive computing will make humans better at accomplishing their goals. Whether it is to ‘find a bad guy’, or to build a better X. In the same way the industrial revolution greatly enhanced productivity (and GDP), cognitive computing will be responsible for the next great leap in human productivity.
In our space, one thing coming in the future is anomaly detection in video. Just as a human can watch a camera feed of a street corner for a few days and get an idea of what is ‘normal’ at all hours of the day, a computer will be able to determine what is normal and alert security professionals when out-of-the-norm behavior is seen.
I also predict that quantity of video being produced by these organizations (ranging from small businesses to enterprises) will continue to grow at a rapid pace - due in part to decreasing cost of surveillance cameras. This will continue to create a massive workload for those responsible for this video, both because of the volume of video and because it lacks any metadata making it searchable.
That makes sense. So, what’s a tall tell that a public safety organization needs an AI / cognitive solution, such as Vergence?
Producing more video than they have manpower to watch. This means most good sized cities in the US. When you do not have enough people to monitor video in real-time it only becomes good after the fact (post-process investigation). By deploying cognitive systems to augment human capabilities video can be used in a more proactive manner to fight crime.
How can a public safety organization ready itself for video AI / cognitive computing solutions?
It’s a good idea to get involved with Federal organizations such as NIST and DHS* S&T. They organize video analytics workshops (which are specifically focused on video analytics in public safety) and participate in larger conferences like SecuredCities. These panels are bringing together researchers developing cutting edge machine learning solutions, customers such as a CIO of a municipality, public safety agencies, and solutions like Vergence to collaborate.
*Jon Vrabel notes that the NIST and DHS workshops coordinate and focus on fostering research, measurement, collaboration and standards among video experts in the public safety community. Both programs are working to bring industry and the research community together with the public safety community to create critical mass in knowledge exchange and R&D.
What is your favorite AI / cognitive computing use case, in public safety?
Favorite use case would be the use of the possibility to use a deep learning model across VergenceNet to help quickly and efficiently catch a kidnapper who has abducted a child and is on the run.
“At the Boston bombing, it took hundreds of analysts days to pour through surveillance video from all of these sources to at first identify the suspects and then to track their exit from the scene. Using a product like Vergence, cognitive computing can allow fewer investigators to follow up on leads much more rapidly at scale and enable more productive collaboration across agencies.” - Jon Vrabel
And what niches do you see AI / cognitive computing public safety solutions benefiting first and why?
The first one that comes to mind is public safety - specifically law enforcement. AI public safety solutions can help efficiently processes massive amounts of live and stored video for high profile events such as a Super Bowl or Presidential Inauguration; help with collecting annotated video for further cognitive computing training; and prevent safety hazards.
The second niche that comes to mind is fire and rescue. They use multiple video sources for things like search and rescue. AI public safety solutions can help autonomously analyze video from a UAS to survey fires, natural disasters like the recent earthquake in Italy, or search for a lost person in the woods by thermal signature would make those processes more efficient.
More generally, any simple task that benefits from a feedback loop could be considered fair game for AI public safety solution applications.
How can we learn more about you and your company?
You can visit us here.
“Cognitive Business” is an interview series featuring prominent figures in the Artificial Intelligence (AI) world. Written by Lolita Taub and written for C-suite and Line-of-Business seeking to address business challenges and goals using the smartest tech.