Building a neural network with the same properties and capacity as the human brain is the holy grail in neuroinformatics. Such a network would not only explain the inner workings of the brain, but would also pave the road for brain-controlled machines such as computers operated by thought and robot limbs for the handicapped. The recently announced IBM Compass project (1) could already provide the necessary capacity for an artificial brain, but to build a truly intelligent network, an accurate model for information processing in the brain is necessary.
Neuroinformaticist Sander Bohte of Centrum Wiskunde & Informatica (CWI) in Amsterdam will present a new model of information processing in the brain on the NIPS 2012 conference in Lake Tahoe (US). In this model, neurons adapt to their circumstances, allowing for a much quicker processing of information. Current models do not take this dynamic behaviour into account.
Most of the workings of the brain are shrouded in mystery. It has the computational power of a hundred modern supercomputers, but the volume of a soda bottle and the energy consumption of a small light bulb. Neuroinformaticists are anxious to discover the secret of such an efficient calculator. According to standard models of neural networks, neurons fire electric pulses called spikes through the network when their inner charge reaches a certain threshold. The faster the neuron receives spikes, the faster it will fire them. This forms the base of calculations in standard neural networks.
According to Bohte, this model is wrong. “Information processing in the brain does not only depend on the rate with which spikes are fired, but also on changes in meaning of this spikes. For instance, when you walk outside, the number of photons that reach your retina increases with a factor of 100 billion. But this doesn’t affect information processing in your brain: after a few moments, you will see just as sharp as indoors.” Bohte includes this dynamical component in his alternative model. The first simulations and tests on experimental data show that the new model’s neural networks are faster and more accurate than current networks, and better resemble the human brain. This model opens the way for a new generation of neural networks that provide a better representation of the brain and are equipped for perform powerful and efficient calculations on problems such as speech recognition and robot movements.
Neuroinformatics is part of CWI’s Life Sciences research. This research uses fundamental theories and tools from mathematics and informatics to make the life sciences more exact and quantitative.
A draft of the paper
About this research
CWI simulates brain activity on video cards