AI defense startups get compute power they need to complete their software - Breaking Defense

A handful of select startup companies have the opportunity to work with Intel’s most advanced chips, processors, and accelerators to supercharge software development in artificial intelligence, machine learning, and analytics. Access to these Intel assets, secure cloud, and expert mentors have helped give these startups the ability to crunch huge amounts of data that provide insights and answer questions posed by national security officials. 

Through the Intel Liftoff program, companies are given access to Intel^® Xeon^® 6 processors with built-in AI acceleration and Intel Gaudi^® 2 and 3 accelerators for large-scale AI tasks, along with the training needed to take full advantage of the chips’ computational power. 

“Liftoff is the right word,” said Steve Orrin, federal security research director and a senior principal engineer with Intel. “It helps get them off the ground without having to do a significant investment in infrastructure nor an investment in having to hire talent that is already pre-trained. We give them access to people who have knowledge and experience to help their engineers and architects craft the right solution.

“Tech mentorships and one-on-one guidance from Intel AI engineers help them understand their workloads and the combination of hardware architectures available to them, and then give them access to that without a major investment on their part. We have the central processing units, graphics processing units, accelerators, and software hosted at Intel, where they can get access to it so they can bring their workload and data, try it out in an environment where they can use all the different architectures and software enablement, and actually run their model, or train their model to do interesting inferencing and to help them accelerate their adoption and deployment.”

Kamiwaza is one of those AI startups that have matured their AI software in the Liftoff program, and as a result are now conducting important work for the Department of Homeland Security Cybersecurity and Infrastructure Security Agency (CISA). 

Kamiwaza’s eponymous AI orchestration engine enables massive dispersed data sets to be rationalized and fused together to achieve mission results.

“A core tenet at Kamiwaza says we want it to be what we call ‘silicon neutral’,” said Luke Norris, CEO and co-founder of Kamiwaza. “We wanted to make sure that our customers could run those outcomes on the right silicon that matched their need – not just cost but the thermal load, speed, a large spectrum of considerations. There is no reason one vendor should fit all because each of these vendors have unique capabilities and structures, and when you can match those capabilities and structures to the outcomes that the customers want then it’s a best-of-all solution. 

“We started with Intel in the Liftoff program to get ourselves situated to figure out how to get our code to work with Intel Xeon 6 and Gaudi 2 and 3 GPUs because that took some technical uplift at the time. We started to get some interest in our work and I was able to go to Intel’s Industrial Solution Builders enablement group and tell them we have a large opportunity with DHS to test all of this data and if it’s positive it’s going to result in, we believe, a level of applications and services we’ll be able to take to the state and fed. They were then able to secure us a couple Gaudi 3 servers in Intel’s cyber cloud and also get us a technical team to make sure we were maximizing the capabilities of the Gaudi 3 chips and cyber cloud.” 

Through Intel’s support and actual production use cases Kamiwaza was able to produce during its time in the Liftoff program, the company was able to demonstrate to CISA that its software had the ability to do an immense amount of data processing and cross correlations on large systems. In addition, leadership at the agency were assured to see that the processing power of the Gaudi 3 GPUs met all of the core benchmarks showing overall wattage usage and impact on the environment. 

Intel also benefits from the partnerships developed under Liftoff, especially in ways that help it improve its own products.

“We’re helping them adopt the technology and drive it,” said Orrin. “For Intel, there’s three benefits. There’s the utilization of our technologies for these novel and startup use cases and services. It’s about how we get our technologies embedded and help them achieve their goals. 

“It’s also collaborative, where we’re learning about the next wave of things that are coming so we can better engineer our products to meet the broader set of ecosystem players where they are. As we learn how they’re leveraging an AI accelerator or a startup is doing interesting graph analytics as part of their AI, are there things we could do in our future hardware to better enable those kinds of use cases. Both sides learn from this exercise.” 

Orrin also noted that Intel has the ability to help these startups gain broader exposure for their work because Intel products are embedded across numerous industries, each with varied use cases and customers. “We can help them get to the right customers and the right partners to accelerate delivery of their products and services,” he said. 

What Kamiwaza produced with Intel’s tools

As part of its DHS work, Kamiwaza is helping to create smart bases, cities, and venues through critical weather correlation and event planning. This is done through an AI orchestration engine that correlates data from 90 years of public and private barometric pressure-related events to predict potential hazardous conditions where life could be threatened. 

“Let me take you through a critical event planning scenario,” offered Norris. “If NOAA comes up with a forecast of a very low barometric pressure event and it’s targeted in a very small region, say a base, an area there’s a training exercise going on, or a large event like an outdoor concert, we can put that into the AI system and it will cross correlate those 90 years of other barometric pressure events that previously happened in that region and it will know all of the impacts of that similar-level event in that area.

“Airports that were shut down, routes and highways that were closed, fallen trees and debris, surge in emergency rooms, you name it. The AI system can actually build a reverse canonical plan based off of historics and based off of the future forecast that we are given. 

“Now you have this massive system that says the last five times an event of this level happened, here was the total impact. Here’s what you should be looking for and here’s how you should proactively plan around that – pre-station troops, pre-station emergency response suppliers, bring in additional resources of X, Y and Z.” 

The ability to use AI to make these types of granular predictions didn’t exist as recently as two-to-three years ago, and will lead to smart bases, smart cities, and smart event planning.