本文關鍵詞： 神經元建模 泊松點過程 通用線性模型 數值梯度下降　出處：《浙江大學》2017年碩士論文　論文類型：學位論文

【摘要】：神經元脈沖信號的建模與預測是神經科學領域的重要研究問題。通過神經元建模來分析脈沖信號的發(fā)放特點,有助于研究學者們更加深刻地理解大腦在執(zhí)行高級認知任務中的工作方式以及神經信息在不同腦區(qū)之間的傳遞方式,從而對大腦的生理特性有一個更好的認識,乃至建立腦機融合的神經假體。本文通過對獼猴PMd腦區(qū)與M1腦區(qū)的神經元脈沖信號構建數理統(tǒng)計模型,來定性、定量地分析二者之間的功能聯系。PMd腦區(qū)與M1腦區(qū)在獼猴的高級認知活動中具有重要作用,對這兩個腦區(qū)的神經元進行建模,有助于研究學者們深入了解兩個腦區(qū)協(xié)同工作的方式細節(jié)。脈沖信號建模存在諸多挑戰(zhàn)。例如,神經元脈沖信號本身包含非常豐富的信號發(fā)放特性,需要模型具備足夠強的表達能力來表征脈沖信號的多樣性;除此之外,神經元所傳送的信息包含在脈沖信號點過程序列之中,需要模型能夠針對脈沖信號的點過程特性充分挖掘特征。本文以泊松通用線性模型為基礎,針對這幾個問題提出了若干改進,全文的貢獻點歸納如下:1.本文借鑒集成學習中混合模型的思想,訓練若干個弱表征能力的子模型,并對其進行混合構成完整模型,從而增強模型整體的表達能力;2.本文通過將泊松通用模型對應的目標函數由最大化似然函數轉化為優(yōu)化Discrete Time Rescaling Kolmogorov Smirnov統(tǒng)計量,借此增強模型對神經元脈沖信號點過程特性的考量;3.本文通過實驗從不同角度驗證所提出的模型的預測能力,實驗結果表明本文模型在擬合優(yōu)度角度能夠保持一個比較突出的結果,同時模型本身維持著一個較好的生物解釋性。
[Abstract]:The modeling and prediction of neuron pulse signal is an important research problem in the field of neuroscience. It is helpful for researchers to understand more deeply how the brain works in performing advanced cognitive tasks and how neural information is transmitted between different brain regions, so as to have a better understanding of the physiological characteristics of the brain. In this paper, a mathematical statistical model of neural pulse signals in the PMd and M1 brain regions of rhesus monkeys was constructed to determine the nature of the neural prosthesis. Quantitative analysis of the functional relationship between the two areas. PMd and M1 brain regions play an important role in the advanced cognitive activities of rhesus monkeys. The neurons in these two brain regions are modeled. It is helpful for researchers to understand the details of how the two brain regions work together. There are many challenges in the modeling of pulse signal. For example, the neuron pulse signal itself contains very rich signaling characteristics. It is necessary for the model to be strong enough to represent the diversity of pulse signals. In addition, the information transmitted by the neuron is contained in the pulse signal point process sequence, which requires that the model can fully mine the characteristics of the point process characteristics of the pulse signal. This paper is based on Poisson's general linear model. The contributions of this paper are summarized as follows: 1. This paper uses the idea of hybrid model in integrated learning to train several sub-models with weak representation ability. A complete model is constructed by mixing it to enhance the expression ability of the model as a whole. 2. In this paper, the objective function corresponding to Poisson's general model is transformed from maximum likelihood function to optimized Discrete Time Rescaling Kolmogorov. Smirnov statistics. The model is used to evaluate the process characteristics of neuron pulse signal points. 3. The prediction ability of the proposed model is verified by experiments from different angles. The experimental results show that the proposed model can maintain a relatively outstanding result in the goodness of fit angle. At the same time, the model itself maintains a better biological interpretation.
【學位授予單位】：浙江大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：Q42;TN911.6

【相似文獻】

相關期刊論文 前10條

1 張寶學,鹿長余;分塊奇異線性模型及其導出的奇異線性模型間的最小范數二次無偏估計等價性研究(英文)[J];應用概率統(tǒng)計;2004年04期

2 張和平;線性模型的比較[J];科學通報;1987年06期

3 張文文;奇異線性模型的估計效率[J];應用數學學報;1995年04期

4 高理峰,劉福升;貝葉斯動態(tài)線性模型的一種實用處理方法[J];山東科技大學學報(自然科學版);2000年04期

5 鄒華國,馬川生,梁t，

本文編號：1465953