Solving Complex Learning Tasks in Brain-Inspired Computers – Eurasia Review
Developing a machine that processes information as efficiently as the human brain is a long-standing research goal towards true artificial intelligence. An interdisciplinary research team from the University of Heidelberg and the University of Bern (Switzerland) led by Dr Mihai Petrovici is tackling this problem using biologically inspired artificial neural networks. Spiked 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 of the main challenges is how to form such complex systems. The German-Swiss research team has now successfully developed and implemented an algorithm that allows such training to be carried out.
Nerve cells (or neurons) in the brain transmit information using short electrical impulses called spikes. These peaks are triggered when a certain stimulus threshold is exceeded. The frequency at which a single neuron produces such spikes and the temporal sequence of the individual spikes are essential for the exchange of information. âThe main difference between biological spike networks and artificial neural networks is that, because they use spike-based information processing, they can solve complex tasks such as recognition and classification of spikes. ‘images with extreme energy efficiency,’ says Julian GÃ¶ltz, doctoral student in Dr Petrovici’s research group.
The human brain and spiked artificial neural networks of similar architecture can only function to their full potential if individual neurons are properly connected to each other. But how can brain-inspired, i.e. neuromorphic, systems be tuned to properly process peak inputs? âThis question is fundamental for the development of powerful artificial networks based on biological models,â emphasizes Laura Kriener, also a member of Dr Petrovici’s research team. Special algorithms are needed to ensure that neurons in a spike neural network fire at the right time. These algorithms adjust the connections between neurons so that the network can perform the required task, such as classifying images with high precision.
The team led by Dr Petrovici has developed such an algorithm. âUsing this approach, we can train spiked neural networks to encode and transmit information exclusively into single spikes. They therefore produce the desired results particularly quickly and efficiently, âexplains Julian GÃ¶ltz. Additionally, the researchers successfully implemented a neural network trained with this algorithm on a physical platform – the BrainScaleS-2 neuromorphic hardware platform developed at the University of Heidelberg.
According to the researchers, the BrainScaleS system processes information up to a thousand times faster than the human brain and requires much less energy than conventional computer systems. It is part of the European Human Brain Project, which integrates technologies such as neuromorphic computing into an open platform called EBRAINS. âHowever, our work is not only interesting for neuromorphic informatics and biologically inspired material. It also recognizes the demand of the scientific community to transfer so-called deep learning approaches to neuroscience and thus to further unveil the secrets of the human brain, âunderlines Dr Petrovici.