Waterloo, Ontario, Canada - March 1, 2018 - At the Intel Lab’s Conference Centre on Intel’s Jones Farm campus just outside of Portland, during the Neuro Inspired Computational Elements (NICE) Workshop, ABR unveiled the first suite of Intel Loihi AI demonstration applications. The NICE conference brings together a global group of leading neuromorphic computing scientists, engineers and corporate representatives.
Announced by Intel’s CEO, Brian Krzanich, this past January during his keynote of the 2018 Consumer Electronics Show in Las Vegas, Intel’s Loihi chip is manufactured in 14nm process technology and delivers leading density and performance for a neuromorphic chip. Learning is supported directly on the chip with programmable learning rules. ABR implemented several of ABR’s spiking learning rules, demonstrating supervised learning and reinforcement learning running directly on the Loihi chip.
Intel showed the first application ever built, a visual object recognition application for Loihi. ABR then unveiled three real-time, online learning demonstration AI applications it created using ABR’s neuromorphic compiler, Nengo, for Loihi. The applications are: (1) a keyword speech recognition app, trained using deep learning and converted to a spiking neural network using ABR’s machine learning to spiking network transfer methods, (2) an expert online reinforcement learning agent that visitors could play in a game of tic-tac-toe, (3) a hybrid neuromorphic robotic controller using online non-linear adaptive control that moves the Kinova Jaco2 arm to track an object detected using a vision system trained with deep learning.
ABR developed the applications for Intel in the span of the eight weeks preceding the workshop. Intel’s Loihi chip performed flawlessly. ABR’s neural modeling and simulation software, Nengo, and Intel’s Loihi provide a complete neuromorphic stack for both low-power, next-generation AI applications development for edge and cloud AI computing, as well as advanced computational neuroscience research. ABR is now working on completing the Nengo backend to fully support Loihi’s features. Loihi will then have access to the full range of components, networks, and applications already developed on Nengo.
ABR ran these applications demonstrating the breadth of uses of the Loihi chip, from inference to reinforcement learning to supervised learning, for the NICE attendees. ABR was honoured to be the first pre-launch collaboration on Loihi and is excited to soon offer all of our collective research and commercial partners the full Nengo software stack to program with Loihi.
ABR’s Nengo is a neuromorphic compiler for building and simulating large-scale neural models for AI, robotics and neuroscience uses. Nengo models are built once and then can run on any of the backends that Nengo supports, including: CPUs, GPUs, MPI and neuromorphic hardware such as Loihi, Spinnaker and BrainDrop. Nengo transforms a functional description of a neural model to a network of spiking or non-spiking neurons. Nengo provides an extensive framework and set of libraries for building applications, reducing the overhead for users to build complex systems. Applications can be constructed with traditional neural network training methods, such as deep learning, or built using ABR’s Neural Engineering Framework (NEF) and Semantic Pointer Architecture (SPA; a library of methods to build cognitive applications).
Nengo on a number of measures is the world’s leading neuromorphic applications tool. Nengo is the result of over 15 years of theory, research, development and optimization by a team of leading computational and theoretical neuroscientists from the University of Waterloo. Nengo runs both from the command line and through an intuitive graphical user interface. Nengo has been used to build neural networks for applications ranging from adaptive robotics control to visual recognition to cognitive tasks such as logical inference and instruction parsing. Nengo was also used to develop the world’s largest functional brain model, Spaun.
More information about Nengo, the Neural Engineering Framework, and Spaun can be found at www.nengo.ai.
Neural networks have become a necessary tool for advanced computing applications. A new generation of computing hardware is emerging designed to efficiently run neural networks, called neuromorphic hardware. These new chips use far less power than convention processors and are able to simulate much larger neural networks much faster, paving the way for efficient embedded processing and further development of advanced AI computing.
Applied Brain Research (ABR) www.appliedbrainresearch.com is the maker of the leading compiler and operating system for neuromorphic computing, Nengo. ABR also builds applications with Nengo, such as AI vision and robotic controllers, for the next major generation of computer chips referred to as neuromorphic hardware.
ABR was launched out of Dr. Chris Eliasmith’s lab, the Centre for
Theoretical Neuroscience at the University of Waterloo, with his
team of leading computational neuroscience and AI researchers. ABR focuses
on the development and commercialization of technology based on theoretical
neuroscience research, providing the neural development and simulation package,
Nengo, as well as neuromorphic solutions to vision, robotics, and
co-CEO & Chairman
Applied Brain Research Inc.