Deliver Your News to the World

Intel Announces Neuromorphic Research Progress


WEBWIRE

What’s New: Today, Intel named academic, government and corporate research groups participating in its Intel Neuromorphic Research Community (INRC) and discussed research progress from the inaugural INRC symposium held in October. The goal of the INRC is to tackle the challenges facing the adoption of neuromorphic architectures for mainstream computing applications. INRC members will use Intel’s Loihi research chip as the architectural focal point for research and development. Intel hopes the findings of this community will drive future improvement of neuromorphic architectures, software and systems, eventually leading to the commercialization of this promising technology.

“While there are many important unsolved neuromorphic computing research problems to explore at all levels of the computing stack, we believe the state of neuromorphic hardware currently leads the state of neuromorphic computing software. We’re confident this network of INRC members will rapidly advance the state of neuromorphic learning algorithms and demonstrate the value of this emerging technology for a wide range of applications.”
–Mike Davies, director of the Neuromorphic Computing Lab, Intel

Who is Participating: Fifty projects have been selected to participate in the INRC. Engaged INRC members will receive access to Intel’s Loihi neuromorphic research chip and software, and are invited to participate in technical symposiums where progress, results and insights will be shared among the community. INRC-supported workshops will offer members an opportunity to learn to develop for Loihi in extended hands-on tutorial sessions and hackathons hosted by Intel Labs researchers and collaborators.

Among the 50 selected projects, teams from 13 universities were selected to receive funding to pursue their research plans. These teams come from a wide range of academic institutions around the world, including University of Bern; University of California, Berkeley; University of California, San Diego; Cornell University; University of Göttingen; TU Graz; Harvard University; TU Munich; Radboud University; University of Tennessee; and Villanova University.

Projects have been scheduled to start over a series of four waves, the first of which began in 2018’s third quarter.

Results So Far: In October, Intel held an inaugural gathering of INRC members in Reykjavik, Iceland. More than 60 researchers attended over five days to discuss research plans, learn about Loihi and meet members of the community. Several presentations from early INRC members announced exciting preliminary progress:

  • Chris Eliasmith of Applied Brain Research Inc. (ABR)* shared early benchmarking resultsevaluating Loihi’s performance running an audio keyword spotting deep network implemented with ABR’s Nengo DL, which runs TensorFlow-trained networks on Loihi. These results show that for real-time streaming data inference applications, Loihi may provide better energy efficiency than conventional architectures by a factor of 2 times to over 50 times, depending on the architecture.
  • Professor Wolfgang Maass of the Institute for Theoretical Computer Science, Technische Universität Graz, discussed his team’s promising discovery of a new class of spiking neural nets that achieve classification accuracies similar to state-of-the-art deep learning models known as long short-term memory (LSTM) networks. LSTMs are commonly used today for speech recognition and natural language processing applications. These new spiking neural networks, named LSNNs, integrate working memory into their operation in a similar manner as LSTMs do, while promising significantly improved efficiency when running on neuromorphic hardware. This work, to be published at the Neural Information Processing Systems conference in December, was developed using a simulator. In collaboration with Intel Labs, Maass’ team is now working on mapping the networks to Loihi. The team shared early accuracy results from the Loihi network, which currently stand within a few percent of the ideal model.
  • Professor Thomas Cleland of Cornell University discussed a set of neuromorphic algorithms for signal restoration and identification in spiking neural networks based on computational principles inspired by the mammalian olfactory system. In work to be published in collaboration with Intel Labs, these algorithms running on Loihi have already shown state-of-the-art learning and classification performance on chemosensor data sets. “These algorithms were derived from mechanistic studies of the mammalian brain’s olfactory circuits, but I anticipate that in generalized form, they will be applicable to a range of similar computational problems such as air and water quality assessment, cancer screening, and genomic expression profiling,” Cleland said.


What Is Neuromorphic Computing: Neuromorphic computing entails nothing less than a bottom-up rethinking of computer architecture. By applying the latest insights from neuroscience, the goal is to create chips that function less like a classical computer and more like a human brain. Neuromorphic chips model how the brain’s neurons communicate and learn, using spikes and plastic synapses that can be modulated based on the timing of events. These chips are designed to self-organize and make decisions in response to learned patterns and associations.

The goal is that one day neuromorphic chips may be able to learn as fast and efficiently as the brain, which still far outperforms today’s most powerful computers. Neuromorphic computing could lead to big advancements in robotics, smart city infrastructure and other applications that require continuous learning and adaptation to evolving, real-world data.

Last year, Intel introduced the Loihi neuromorphic test chip, a first-of-its-kind research chip with an unprecedented combination of neuromorphic features, efficiency, scale and on-chip learning capabilities. Loihi serves as the architectural foundation for the INRC program. Intel provides INRC members with access to this leading neuromorphic chip to accelerate progress in this field of research.

What is Next: Intel has released early versions of its software development kit for Loihi, named Nx SDK, to engaged INRC members. Researchers may remotely log in to Intel’s neuromorphic cloud service to access Loihi hardware and Nx SDK to develop their algorithms, software and applications. Additionally, Intel has supported Applied Brain Research to port its Nengo software framework to work with Loihi. Nengo is freely available today for research use.

Loihi hardware has been made available to select INRC members for research in domains such as robotics that require direct access to hardware. These systems include a USB form factor code-named “Kapoho Bay.” In addition to providing a USB interface to Loihi, Kapoho Bay offers an event-driven hardware interface to the DAVIS 240C DVS silicon retina camera available from iniVation*, among other peripherals.

Next year, Intel and INRC members expect to contribute much of the enabling software and research results to the public domain in the form of publications and open source software. INRC membership is expected to steadily grow, and as the foundational algorithms and SDK components mature, Intel foresees an increasing project focus on real-world applications, ultimately leading to the commercialization of neuromorphic technology.

How to Get Involved: Neuroscientists, computational scientists and machine learning researchers interested in participating in the INRC and developing for Loihi are encouraged to email inrc_interest@intel.com for more information.

Additionally, Intel’s Neuromorphic Computing Lab will support full-day tutorials on Loihi’s systems and software at two upcoming events: at the 2019 Riken International Workshop on Neuromorphic Computing in Kobe, Japan, on March 13, and at the 2019 Neuro Inspired Computing Elements (NICE) Workshop in Albany, New York, on March 29. The tutorials will be open to all registered attendees of these workshops.


( Press Release Image: https://photos.webwire.com/prmedia/6/232637/232637-1.jpg )


WebWireID232637





This news content was configured by WebWire editorial staff. Linking is permitted.

News Release Distribution and Press Release Distribution Services Provided by WebWire.