AI in HPC on Wall Street: Getting Going Is the Hard Part 

IT managers in the financial services industry are fully aware that high performance data analytics, AI and deep learning technologies are plentiful and available, and they know that some FSI companies are already deriving advantage from them. But they hesitate to take the plunge into AI/DL themselves because of sticker shock ($200,000 for an initial implementation?), technology unfamiliarity, dearth of trained staff and lack of appropriate server infrastructure.

This might not be the case at the high end of FSI, organizations with 100-plus data scientists on hand and deep IT budgets to buy the infrastructure and consulting services to execute optimized AI strategies. But at many mid-sized and smaller FSI companies, more people are talking about AI than are actually implementing it, more are looking for a way to initiate AI than are actually taking the first steps.

Helping organizations begin their AI/DL journeys is a theme of the keynote presentation to be delivered at next week’s HPC for Wall Street – Cloud and Data Centers conference in New York by Lacee McGee, HPE’s worldwide senior product manager for financial services industries. Her address also will highlight the growing need for code modernization: updating custom FSI software and algorithms to take advantage of new processor architectures, along with cloud implementation strategies.

HPE's Lacee McGee

“There hasn’t emerged a front-runner AI/DL solution,” McGee told EnterpriseTech. “Customers are hungry for solutions, but they don’t know what they entail. What we see as they start to tinker around with it, they’re not ready to invest a significant amount of money needed for an optimal AI/DL set up. They want to play around on something familiar to them – a 2u standard link form factor box that they already have for the email server or some other purpose. They hope if they start on something comfortable and general-purpose then maybe they can begin down that path.”

Her advice on an AI starter strategy: Go with what you know.

“So the opportunity is how can we give them little bits of AI and deep learning, packaging that starts on hardware they’re familiar with?” McGee said. “Then as they grow and get more comfortable they can make smarter decisions and transition to those optimal solutions. They’re looking for comfort with both technology and the price point.”

This approach contracts with most AI and DL systems currently available, she said. “The systems I’ve seen are very specific, so it’s not something that if you tried and it didn’t work you could repurpose for a standard workload, like enterprise Sharepoint or your email server. So the more we can deliver flavors of AI and DL and guidance on the journey, starting at a more cost-effective price with a more uniform architecture, that should ramp up adoption more quickly.”

McGee also will discuss appropriate workloads for initial AI implementations, beginning with anomaly detection, which is inherently well-suited for FSI because the enormous volumes of data generated that can be used to train DL systems. “Due to regulations and making smarter market trades in the HFT section, financial services generate so much data that they’re required to store, they have an advantage over other industries that have data but don’t hold onto it. Our customers are holding onto it, so they might as well leverage it and make smarter decisions in the anomaly detection space.”

Another main element of her address to the FSI crowd: at-scale cloud implementation.

“It’s involved with everything, whether HPC or not in the financial services industry, ” McGee said, “because of fear of cybersecurity problems, or not being comfortable with new regulations – not knowing if cloud providers are tied in with new regulations will undermine adoption of cloud, whether it’s public, on-premises or hybrid cloud.”

McGee said cybersecurity concerns are the biggest barrier to cloud computing in FSI. “There’s so much going on with cybersecurity, there’s slower adoption than in other industries,” she said. “It’s a very small percentage of workloads customers are actively transitioning to coudl platforms. They see the benefit (from cloud computing) in saving money and accessing data. But the cost of something going wrong is so great that we’re farther behind than other industry verticals.”

McGee will also discuss incorporation of alternative processor architectures to speed up workflows.

“It’s always been there, say with low latency high frequency trading,” she said, “how do I reduce my time, how do I take advantage of emerging accelerated technologies?” She explained that simply dropping a state-of-the-art GPU into the processor mix will improve performance, the improvement will probably fall short of expectations. This is because algorithms and codes need to be optimized to fully utilize new architectures, McGee said. Large FSIs with big IT staffs can take on those challenges in-house, but smaller companies in the industry increasingly are engaging code modernization consulting firms to help them update their proprietary software.