Probability Processors For Accelerating Statistical Inference And Learning
Probability Processors are a new class of silicon architectures being developed by Analog Devices Lyric Labs, specifically for accelerating statistical inference and machine learning applications. Our initial silicon implementations have demonstrated orders of magnitude energy and performance wins over conventional processors for solving a range of probabilistic graphical models.
In order to make it easier for a wide range of developers and researchers to create machine learning applications in their domains of interest, over the past ten years we have created a growing Bayesian tool stack that includes open source probabilistic modeling languages and specialized "compilers." We are working with the probabilistic programming community grow and enhance the range of applications that we can compile to efficient hardware architectures.
There are a wide range of applications for probability processors. We will show one or two, and hope that we will have the opportunity to discuss more with the audience.
About Ben Vigoda and Lyric Labs, Analog Devices Inc. :
The Analog Devices corporate research labs, located in Kendall Square, grew from the acquisition by Analog Devices of Lyric Semiconductor, a startup company that Ben Vigoda co-founded to focus on the business and science of implementing statistical machine learning and semantic signal processing natively in semiconductor circuits. As CEO, Vigoda raised over $25M in venture capital funding and government research contracts while also contributing to the company's technology development. Lyric was selected as one of the 50 Most Innovative Companies by Technology Review in 2010. The company's work on probability processors has been featured in the Wall Street Journal, New York Times, Wired Magazine and Scientific American and was #1 on slashdot for one exciting day.
Vigoda developed the technical foundations for Lyric during his PhD work with Prof. Neil Gershenfeld at the Center for Bits and Atoms in the MIT Media Lab, and subsequently while a Research Scientist at Mitsubishi Research Labs. While a PhD student, he learned a lot about how to create a startup as part of the process of winning the MIT 50k Entrepreneurship Competition and Harvard New Venture Competition. He earned his undergraduate degree in physics from Swarthmore College and has worked at the Santa Fe Institute on alternative models of computation and at Hewlett Packard Labs, where he helped transfer academic research projects to product divisions, including a toner level sensor that made its way into every HP Laserjet printer. He has also worked on some other interesting projects, including programming the automated design of DNA-based nano-tile structures, building a virtual juggling system that toured for years with the Flying Karamazov Brothers, and co-founding Design That Matters, a not-for-profit that collaborates with universities and volunteer engineers to design new products and services for the poor in developing countries.