The Pattern Recognition Lab

Coupling Machine Learning to Brain-Inspired Computers

SEATTLE, WA --(Marketwired - September 28, 2016) - "The scientific method traditionally begins with a hypothesis, which is then tested against data. And we've learned that creating the right hypothesis is the single largest contributor to major scientific advances. Today, powerful new 'brain-inspired' computing capabilities are accelerating a 'data science' experimental method -- a method that detects patterns in data as a critical first step in generating a hypothesis," explained Mark Anderson, CEO of The Strategic News Service and the SNS Future in Review (FiRe) Conference.

"Pattern recognition is a mode of epistemology, a way of knowing," says University of California San Diego's Larry Smarr, director of the California Institute for Telecommunications and Information Technology (Calit2). "It's taking the same data that's available to everyone and trying to let the data talk to you instead of putting your preconceived notions onto it."

Both machine learning (ML) techniques and novel computer architectures are rapidly developing, so a major challenge is how to optimize a variety of ML algorithms on different architectures and discover which are fastest with the most energy efficiency for specific applications across a wide range of disciplines. Furthermore, there must be flexibility to both process massive static arrays of data as well as myriad flows of data -- and find never-before-seen patterns in both.

To explore these trade-offs, the Strategic News Service and Calit2 have created a Pattern Recognition Laboratory (PRLab), housed in Calit2's Qualcomm Institute at UC San Diego. The PRLab is in the early stages of building a "garden of architectures" capable of performing massive amounts of high-speed processing without consuming as much power as traditional chips.

UC San Diego Professor Ken Kreutz-Delgado, a long-time member of the Electrical and Computer Engineering Department, is the PRLab's first director. Kreutz-Delgado is taking a broad view of the disciplines to which pattern-recognition computing can be usefully applied.

"It isn't just science and engineering problems, but also extends to arenas in sociology, politics, economics… any discipline where data can be collected and analyzed with models from the bottom up," said Kreutz-Delgado.

Besides powerful traditional von Neumann architectures such as shared-memory multi-core and graphics processing units (GPUs), the PRLab has acquired non-von Neumann architectures such as high density Field Programmable Gate Arrays (FPGAs), IBM's TrueNorth neuromorphic processor, and KnuEdge's LambdaFabric™ neural computing systems.