Mechanism Study Of Neural Coding Based On Symbolic Dynamics

Computational neuroscience plays an increasingly important role in the recent30years. It has achieved some fruitful research results with the methods of the nonlinear dynamics. As the main carrier of the neural information expression, neural coding is the foundation of the brain research. It is also an important part in the road of artificial intelligence development. Due to the complexity of the nervous system and the intrinsic randomness of the spike trains, there are many disputes about how the spike trains encode the external message. Whether the rate coding or temporal coding, each coding theory has some shortcomings in explaining the mechanism of rapid and precise neural responses. Therefore, it is significant to develop a high resolution, keeping monotonic with the stimulus and fast response spike coding theory in neural information study.In this dissertation, the dynamical behavior of the spike trains is investigated with the principle of circle maps and symbolic dynamics. First, we take the Hodgkin-Huxley neuron model and neural network as research objects. And then we discuss the status of spike train pattern under different frequencies of the external current stimulus or different synapse coupling strengths. With analyzing the relationship between the neural action potential and the parameters of the current stimulus, a new neural coding theory is proposed based on the orderable of the symbol trains. We named it as symbol orderable coding theory. The main results of this dissertation are summarized as following:Firstly, simulation results show that in contrast to the other neural coding, the major advantage of the symbol orderable coding theory is the monotonic relationship between the stimulus strength and the symbol sequences. In addtion, the symbol orderable coding theory has two other advantages. They are shorter response time and higher sensitivity. When the length of the symbol sequences more than ten, it can distinguish the time differences less than20microseconds. The dissertation takes the sound localization neural mode as an example. Comparing with the rate coding, the symbol orderable code needs less time to distinguish the same resolution interaural time difference and interaural level difference singals. This illustrates that the symbol orderable coding theory is more suitable for studying the mechanism of fast and precise neural react.Secondly, this dissertation studies the partition and the ording rules of the symbolic dynamics under aperiodically driven. With these studies, the symbol orderable coding can analyze the action potential not only in the periodically driven neuron model, but also in the aperiodically driven neuron or neural network models. It takes the application of the symbol orerable coding more in line with the physiological characteristic of neural system.Thirdly, an instance of neural coding application for parameter estimation in neural system implied that the symbol orderable coding theory might be a potential and useful method in pattern recognition. In order to estimate the system parameter precisely, almost all the previous approaches need the high sensitive measurement as the premise. But with the symbol orderable coding theory, the system parameter can be estimated precisely without the high sensitive measurements. The advantage reduces the measurement requirement and reflects robust to the system noise. Therefore, the parameter estimation method is worth to be popularized in artificial intelligence study.In conclusion, this dissertation proposes a novel neural coding theory based on the orderable of symbolic sequences. This neural coding theory can analyses the intrinsic randomness neural spike trains quantitatively. And it has significant advantages in revealing the pathogenesis of nerve disease and developing the artificial intelligence.