Neuromorphic Computing- QUBO Solver Visualization, Satellite Scheduling
Developed with Neuromorphic Computing researchers at Intel Labs, this interactive visualization highlighted a real-world satellite-scheduling QUBO problem (Quadratic Unconstrained Binary Optimization.) The visualization rendered in real-time, while graphing data from a state-of-the-art CPU QUBO solver, along with Intel Lab's Neuromorphic Computing QUBO solver running on an Intel Loihi2 Neuromorphic processor. ref. Solving QUBO on the Loihi 2 Neuromorphic Processor
Situation: Intel's Neuromorphic Computing team was making steady gains in performance, but needed a more compelling technical storytelling demo for a major upcoming event (Intel Innovation.)
Task: As the events demo TME lead for Intel Labs, I needed to work with the Neuromorphic Computing team to identify opportunities for live demos and problem visualizations/explainers that would be both captivating for technical and non-technical audiences, while also clearly and accurately explaining the technology and it's benefits. The demo would be presented by Neuromorphic Computing researchers and serve as a storytelling attractor, and a technical visual aid.
Action/Process: I met with Neuromorphic Computing Lab researchers to identify the most promising areas of current research, and how they overlap with real-world problems that we could build into a demo. It was critical that the demo be grounded in a real example, ideally with running neuromorphic computing hardware feeding a data visualization.
We quickly settled on constraint optimization and the researchers' work with QUBO solvers, and landed on a satellite-scheduling problem as an opportunity for a compelling and intuitive visualization, as well as a running hardware demonstration and data graphing.
I worked with a 3D artist, designer, and Intel Labs researchers to build this demo for large 4K screens at Intel Innovation and future events. The demo consisted of a technical storytelling element, and a live data visualization, and was designed with the ability to run on live results from the neuromorphic hardware and cached data from prior runs.
This visualization had three interactive components, (below) a dynamic graph of the Spiking Neural Network (SNN) gradient descent (bottom left,) a plot of EDP performance vs capture requests on both a QUBO CPU solver and Neuromorphic solver (right,) and a live rendered satellite over the earth with target waypoints appearing/dissapearing as the number of requests changed. Due to time limitations on a show floor, the graphs in this demo referenced a table of cached hardware results, while we had a Loihi2 development platform running the QUBO problems on the counter with a separate terminal display.
Impact: This demo was a communications success for Intel Labs researchers. In addition to successfully anchoring the Neuromorphic Computing outreach on the show floor of several major events, the visualization was used by the director of Intel Labs, Rich Uhlig in his keynote presentation at Intel Vision.
Below, a researcher presents this demo to the Intel CEO on the show floor of Intel Innovation 2023.