Final Report Summary - SPIKEHEAR (Spiking neural models of auditory perception)
Neurons compute mainly with action potentials or “spikes”, which are stereotypical electrical impulses. Over the last century, the operating function of neurons has been mainly described in terms of firing rates, with the timing of spikes bearing little information. This classical point of view has led to a considerable number of developments in computing, from the perceptron to modern artificial neural network theories for pattern recognition. However, recent experimental evidence and theoretical studies show that the relative spike timing of inputs has an important effect both on computation and learning in neurons, which has triggered considerable interest for spiking neuron models in computational neuroscience.
This project has developed a theory of spike-based computation based on selective synchronization of neurons, in the context of sensory systems. In this framework, synchrony between neurons signal the presence of a specific invariant (or “law”) in the sensory signals. This theory has been applied to two problems faced by the auditory system: sound localization and pitch perception. In sound localization, we have developed spiking neuron model based on our theory, which was shown to be accurate in realistic acoustical conditions. The model produced specific predictions about the relationship between acoustical properties of natural environments and physiological properties of spatially-tuned neurons, which we tested successfully. We have also developed a functional spiking neuron model that estimates the fundamental frequency of sounds. The model instantiates a new theory of pitch perception, according to which pitch is the perceptual correlate of the regularity structure of the basilar membrane vibration. A key prediction of the theory, which we successfully demonstrated, is that low-frequency pure tones sound lower in pitch when sound level is increased.
Finally, a technological product of the project is the development of a general-purpose spiking neural network simulator, the Brian simulator (http://briansimulator.org) which is open source and freely available. The simulator allows anyone to simulate our models but also any kind of spiking neural network model. It is used very broadly in the computational neuroscience community.
This project has developed a theory of spike-based computation based on selective synchronization of neurons, in the context of sensory systems. In this framework, synchrony between neurons signal the presence of a specific invariant (or “law”) in the sensory signals. This theory has been applied to two problems faced by the auditory system: sound localization and pitch perception. In sound localization, we have developed spiking neuron model based on our theory, which was shown to be accurate in realistic acoustical conditions. The model produced specific predictions about the relationship between acoustical properties of natural environments and physiological properties of spatially-tuned neurons, which we tested successfully. We have also developed a functional spiking neuron model that estimates the fundamental frequency of sounds. The model instantiates a new theory of pitch perception, according to which pitch is the perceptual correlate of the regularity structure of the basilar membrane vibration. A key prediction of the theory, which we successfully demonstrated, is that low-frequency pure tones sound lower in pitch when sound level is increased.
Finally, a technological product of the project is the development of a general-purpose spiking neural network simulator, the Brian simulator (http://briansimulator.org) which is open source and freely available. The simulator allows anyone to simulate our models but also any kind of spiking neural network model. It is used very broadly in the computational neuroscience community.