Computers learn to learn

Researchers from Heidelberg University and TU Dresden, together with Intel Corporation, will reveal three new neuromorphic chips during the NICE Workshop 2018 in the USA. These chips have an extraordinary ability: They are able to mimic important aspects of biological brains by being energy efficient, resilient and able to learn. These chips promise to have a major impact on the future of artificial intelligence. Computers are many times faster than humans in solving arithmetical problems, yet they have thus far been no match when it comes to the analytic ability of the brain. Up until now, computers have not been able to continually learn and can therefore not improve themselves. The two European chips were developed in close collaboration with neuroscientists as part of the Human Brain Project of the European Union. NICE 2018 will be held from 27 February until 1 March on the Intel Campus in Hillsboro/Oregon.

Dr Johannes Schlemmel from the Kirchhoff Institute for Physics at Heidelberg University will present prototypes of the new BrainScaleS chip. BrainScaleS has a mixed analogue and digital design and works 1,000 to 10,000 times faster than real time. The second generation neuromorphic BrainScaleS chip has freely programmable on-chip learning functions as well as an analogue hardware model of complex neurons with active dendritic trees, which – based on nerve cells – are especially valuable for reproducing the continual process of learning.

Dr Sebastian Höppner from the Chair of Highly-Parallel VLSI Systems and Neuro-Microelectronics at TU Dresden will present prototypes of the new SpiNNaker chip. The second generation of SpiNNaker is based on many-core architecture, which was developed by Prof. Dr. Steve Furber at the University of Manchester. Prof. Furber is a pioneer of so-called ARM architecture, which today is found in every smartphone. The new SpiNNaker chip is based on this technology; a multitude of processor cores has been integrated on this chip. A single chip contains 144 ARM Cortex M4 cores with innovative power management for very efficient energy usage. The SpiNNaker chip delivers computational power of 36 billion instructions per second, per watt. Its primary application with be for real-time simulation of multi-scale brain models.

Intel Corporation will also unveil the technical details of their recently announced Loihi research chip. This new processor technology contains a highly developed command structure for neural networks from “firing” neurons, as well as microcode-programmable learning rules. Loihi supports a range of on-chip learning models from supervised and unsupervised learning to reinforcement-based learning processes.

All three new chips will be presented as hardware prototypes. Experts are available for potential users to discuss properties and applications. Additional presentations during the workshop will deal with the theories and algorithms of neuromorphic chips. The acronym NICE stands for “Neuro Inspired Computational Elements”. The NICE workshop takes place annually and draws together researchers from Europe and the USA who are developing new computer architectures based on neural systems. The workshop series was initiated in 2013 by the physicist Prof. Dr Karlheinz Meier.

Information for journalists:

Prof. Dr Karlheinz Meier
Heidelberg University
Kirchhoff Institute for Physics
Phone +49 6221 54-9831
meierk@kip.uni-heidelberg.de

Prof. Dr Christian Mayr
TU Dresden
Faculty of Electrical and Computer Engineering
Chair of Highly-Parallel VLSI Systems and Neuro-Microelectronics
Phone +49 351 463-42392

Further information on the internet:

NICE 2018 – http://niceworkshop.org/nice-2018-agenda