idw – Informationsdienst Wissenschaft
Nachrichten, Termine, Experten
Nachrichten, Termine, Experten
Spiking neural networks, which mimic the structure and function of a natural nervous system, are powerful, fast, and energy-efficient. One key challenge is how to train such complex systems. An interdisciplinary research team at Heidelberg University and the University of Bern (Switzerland) has now developed and successfully implemented an algorithm that achieves such training. It can be used to train spiking neural networks to solve complex tasks with extreme energy efficiency.
Press Release
Heidelberg, 29 October 2021
Solving Complex Learning Tasks in Brain-Inspired Computers
Researchers from Heidelberg and Bern develop a new training approach for spiking neural networks
Developing a machine that processes information as efficiently as the human brain has been a long-standing research goal towards true artificial intelligence. An interdisciplinary research team at Heidelberg University and the University of Bern (Switzerland) led by Dr Mihai Petrovici is tackling this problem with the help of biologically-inspired artificial neural networks. Spiking neural networks, which mimic the structure and function of a natural nervous system, represent promising candidates because they are powerful, fast, and energy-efficient. One key challenge is how to train such complex systems. The German-Swiss research team has now developed and successfully implemented an algorithm that achieves such training.
The nerve cells (or neurons) in the brain transmit information using short electrical pulses known as spikes. These spikes are triggered when a certain stimulus threshold is exceeded. Both the frequency with which a single neuron produces such spikes and the temporal sequence of the individual spikes are critical for the exchange of information. “The main difference of biological spiking networks to artificial neural networks is that, because they are using spike-based information processing, they can solve complex tasks such as image recognition and classification with extreme energy efficiency,” states Julian Göltz, a doctoral candidate in Dr Petrovici’s research group.
Both the human brain and the architecturally similar artificial spiking neural networks can only perform at their full potential if the individual neurons are properly connected to one another. But how can brain-inspired – that is, neuromorphic – systems be adjusted to process spiking input correctly? “This question is fundamental for the development of powerful artificial networks based on biological models,” stresses Laura Kriener, also a member of Dr Petrovici’s research team. Special algorithms are required to guarantee that the neurons in a spiking neural network fire at the correct time. These algorithms adjust the connections between the neurons so that the network can perform the required task, such as classifying images with high precision.
The team under the direction of Dr Petrovici developed just such an algorithm. “Using this approach, we can train spiking neural networks to code and transmit information exclusively in single spikes. They thereby produce the desired results especially quickly and efficiently,” explains Julian Göltz. Moreover, the researchers succeeded in implementing a neural network trained with this algorithm on a physical platform – the BrainScaleS-2 neuromorphic hardware platform developed at Heidelberg University.
According to the researchers, the BrainScaleS system processes information up to a thousand times faster than the human brain and needs far less energy than conventional computer systems. It is part of the European Human Brain Project, which integrates technologies like neuromorphic computing into an open platform called EBRAINS. “However, our work is not only interesting for neuromorphic computing and biologically inspired hardware. It also acknowledges the demand from the scientific community to transfer so-called Deep Learning approaches to neuroscience and thereby further unveil the secrets of the human brain,” emphasises Dr Petrovici.
The research was funded by the Manfred Stärk Foundation and the Human Brain Project – one of three European flagship initiatives in Future and Emerging Technologies supported under the European Union’s Horizon 2020 Framework Programme. The research results were published in the journal “Nature Machine Intelligence”.
Contact:
Heidelberg University
Communications and Marketing
Press Office, phone +49 6221 54-2311
presse@rektorat.uni-heidelberg.de
Julian Göltz
Kirchhoff Institute for Physics
Phone +49 6221 54-9159
julian.goeltz@kip.uni-heidelberg.de
J. Göltz, L. Kriener, A. Baumbach, S. Billaudelle, O. Breitwieser, B. Cramer, D. Dold, A. F. Kungl, W. Senn, J. Schemmel, K. Meier, M. A. Petrovici: Fast and energy-efficient neuromorphic deep learning with first-spike times, Nature Machine Intelligence (17 September 2021), https://doi.org/10.1038/s42256-021-00388-x
http://www.kip.uni-heidelberg.de/vision
https://physio.unibe.ch/~petrovici/group/index.aspx
https://ebrains.eu/services/brain-inspired-technologies
Illustration of the on-chip classification process with the Yin-Yang dataset. Each symbol represents ...
Göltz and Kriener et al. (Heidelberg / Bern)
Criteria of this press release:
Journalists, Scientists and scholars
Information technology
transregional, national
Research results, Scientific Publications
English
Illustration of the on-chip classification process with the Yin-Yang dataset. Each symbol represents ...
Göltz and Kriener et al. (Heidelberg / Bern)
You can combine search terms with and, or and/or not, e.g. Philo not logy.
You can use brackets to separate combinations from each other, e.g. (Philo not logy) or (Psycho and logy).
Coherent groups of words will be located as complete phrases if you put them into quotation marks, e.g. “Federal Republic of Germany”.
You can also use the advanced search without entering search terms. It will then follow the criteria you have selected (e.g. country or subject area).
If you have not selected any criteria in a given category, the entire category will be searched (e.g. all subject areas or all countries).
