Booz Allen Gives Government a Deep Learning Edge
The latest breakthroughs in deep learning technology have emanated from places like Silicon Valley and Toronto, where Turing Award-winner Geoffrey Hinton did his seminal work. But deep learning applications today are finding their way to Washington D.C., often through the assistance of government contractors like Booz Allen Hamilton (BAH).
Josh Elliot, the principal and director of AI for Booz Allen Hamilton, says the company is actively engaged in various deep learning projects within various branches of the Federal Government.
“We’re absolutely applying [deep learning],” Elliot told Datanami at Nvidia‘s GPU Technology Conference (GTC) in San Jose, California. “We have probably 60 or more [projects] right now, today, active in deep learning, machine learning, and artificial intelligence types of projects across the Federal Government right now.”
The projects span everything from the healthcare domain and predictive maintenance to helping our veterans and supporting DOD and intelligence community missions. Cyber security is a major use case for AI technologies in BAH’s government contracts, as well as detecting fraud, waste, and abuse.
While there’s a lot of interest in AI, it’s necessary to separate what’s possible with today’s technology from what is not. To that extent, one of BAH’s roles is to “demystify” AI and DL technologies for its government clients.
“There’s big hype and a draw to some of these sophisticated technologies, and you need to understand the tradeoffs and balances of where to apply what and when to make sure it’s a successful adoption,” he said. “We talk a lot about AI and DL and where it applies. We talk about use cases with different clients. There is opportunity for applying these technologies, but I think it needs to be balanced with what the mission is, what the infrastructure can support, what data exits, and the talent assets that you have.”
Successful DL implementations typically utilize very large data sets and expensive GPU infrastructure, which is beyond the scope and capability of some of its government clients.
“You don’t want to put something in that’s too sophisticated for the maturity of your infrastructure to support long term,” Elliot said. “The last thing you want to do is put a solution out there that doesn’t have the infrastructure, because you disenfranchise people pretty quickly in terms of the adoption. You exit that much harder.”
BAH has sophisticated deep learning implementations with the government, but the learning curve is still quite steep for multiple aspects of this technology, Elliot said.
“This technology is still pretty nascent, and specifically with the federal clients, there’s a lot of R&D that still needs to take place and a lot of proof of concepts and prototypes that need to be played out,” he said. “But one of the most important things that we do in thinking through the design and implementation of these technologies in designing and implementing a machine learning or deep learning algorithm in a development environment is extremely different when you try to operationalize that into product.”
The McLean, Virginia-based company and its federal clients are wrestling with some of the same architectural issues that their cohorts in the private sector are dealing with, specifically as it relates to the collection, storage, and processing of data in a geographically distributed environment. That includes being aware of the opportunities and challenges of extending one’s compute environment from the edge to the core, particularly with the advent of fast 5G networks.
“It’s very different deploying these into a tactical environment,” Elliot said. “Think about mobile devices or devices out in the field. You’ve got to design the algorithm different and you need to understand the data and how it’s going to be used, where that processing needs to occur. How much data do you actually need to pull off the sensor, or can you do some of that processing for instance on the sensors and then send back the pertinent information into an enterprise type of deployment? So there’s a lot of nuances around the infrastructure, the network.”
BAH partners with NVidia not only on the hardware aspects of training deep neural networks, but also through programs like NVidia’s Deep Learning Institute to educate the next generation of data scientists, Elliot said.
Along with all the other companies engaging in data science and AI, BAH is chasing a finite number of skilled data science and AI practitioners. To that end, the GTC19 conference, which attracted upwards of 8,000 people to the San Jose McEnery Convention Center last week, was a good place to be.
“This is a tech conference. It brings lot of the brightest minds in the world,” Eilliot said. “I’m grateful for NVida for hosting these types of events because it gives us opportunity to learn, but also to identify potential partners, to identify potential candidates for recruitment.”
Some firms in the Silicon Valley such as Google have declined to bid on government contracts over what they see as overreach by the government. Elliot says he respects every company’s right to make its own decisions. At the same time, he views the nature of some of BAH’s work with the federal government to be a plus for a certain type of patriotic citizen.
“I think one of the value propositions that we offer individuals is a diversity of mission and supporting the purpose of serving the citizens and the war fighters or the safety specialist,” he said. “I think that is an attractive option for some people.”