Analysis on High-Speed Computation Mechanisms of the Brain Based on Impulse-Processing Neural Networks

基于生理数据的脉冲处理神经群行为分析表明，大脑神经群可以检测到毫秒级的激活时间差异，并且这种能力对噪声具有足够的鲁棒性。这暗示了一种可能性，即神经核活动的时间可能控制大脑在几毫秒范围内的并行处理，以实现高速计算。利用脉冲处理神经群模型，提出了新皮质-丘脑系统奖励学习机制的数学模型。该模型描述了单个神经元的生理参数的功能，即大脑的短期记忆需要持续多长时间才能完成奖励介导的学习，学习完成需要重复多少次学习。这些函数根据单个神经元的生理数据提供了对上述系统行为的理论估计，这使我们能够通过检查估计与系统实际行为的一致性来测试所提出的模型。从耳到皮层的听觉通路在解剖学上被划分为5个部分，根据这5个部分的生理数据，构建了听觉通路的数学模型，研究了听觉通路的行为。对模型行为的计算机模拟表明，该模型不仅能够识别连续语音中的孤立音节和连接词，而且能够描述真实听觉系统中单个神经元的典型反应。这支持了为构建我们的模型而引入的假设。

Analysis of behavior of impulse-processing neuropopulations based on physiological data has shown that neuropopulations of the brain can detect millisecond time differences in their activations and that the capacity has sufficient robustness against noises. This suggests a possibility that timing of activities of neural nuclei may control parallel processing of the brain in the several millisecond range to achieve high-speed computation.A mathematical model of reward-mediated learning mechanisms of the neocortico-thalamic system was proposed using the model of impulse-processing neuropopulations. The model describes as functions of physiological parameters of single neurons how long the short-memory of the brain must last to perform the reward-mediated learning how many repetitions of learning are required for learning accomplishment. These functions provide theoretical estimation on the behavior of the above system according to physiological data on single neurons, which enables us to test the proposed model by examinig agreement of the estimation with real behavior of the system.Behavior of the auditory pathway from ear to cortex, which is anatomically divided into five divisions, was investigated by constructing a mathematical model of the pathway according to physiological data on the five divisions. Computer simulation of the model behavior has shown,that the proposed model provides not only the capacity of recognizing isolated syllables and connected words in continuous speech but also describes typical response of single neurons observed in the real auditory system. This supports assumptions introduced to construct our model.