臺灣博碩士論文加值系統

棘波序列度量法(Spike Train Metrics)之目的，是計算出兩筆棘波序列的差異性，而此差異性可表示兩棘波序列是否為相同刺激反應而來；由於棘波序列度量法只能針對兩個棘波序列間的判別其差異，在同時面對大量的棘波序列時，會是一個難以解決的問題；在此我們使用值譜分群(Spectral Clustering Algorithm)的方式，計算出棘波序列彼此的差異性，利用棘波序列之間相對的距離，我們從中找出分群的憑據，由此我們可判別未知的棘波序列為何者刺激而產生。
雖然實驗中已盡量排除各種影響棘波序列的狀況，但實際的棘波序列資料仍不易分析，且無正確答案可當作參考；所以本論文中利用已知不同反應的棘波序列當作樣本，依據棘波刪除與棘波位移兩種方式，產生數組模擬棘波序列的數據，透過我們提出的分群方式，觀察模擬棘波刪除或位移的程度、棘波序列度量法中時間精準度的參數不同時，是否還能區別已知不同反應的棘波序列。
由數據結果我們可發現，Victor與van Rossum各自提出的棘波序列度量法，在比對棘波序列上各有其限制；棘波個數的多寡，也會影響區別棘波序列之差異性；時間精準度的選取值差異，除了影響比對棘波序列上的判斷外，在模擬棘波刪除、模擬棘波位移兩種方式，也有其不同的影響。

The purpose of spike train metrics is to realize the discrimination of two spike trains, and the difference of the distance may measure the dissimilarity and identify whether these two spike trains come from the same stimulus or not. Since the spike train metrics is used to differentiate two spike trains, it suffers from a difficulty in how to measure the dissimilarity for a large number of spike trains at the same time. In this research, we use the spectral clustering algorithm to discover the similarity among several spike trains directly. By doing so, we expect to understand which stimulus is corresponding to an unknown spike train. This research studies two spike train metrics, one is proposed by Victor and the other is introduced by van Rossum. To evaluate their performance, we generate simulated spike trains to reduce the complexity of real data. Two cases are considered in the simulation, case 1 concerns the effect of spike deletion (insertion), and case 2 focus on the spike shifting. This research explores the capabilities of two spike train metrics through the use of the spectral clustering algorithm with different parameter settings. The study conducts several experiments and analyzes the strengths and limitations of these two metrics. According to the experimental results, we find strong evidence for the number effect of spikes. We also discover how the parameter settings of time accuracy influence the clustering on the spike deletion (insertion) and spike shifting. Finally, we conclude that Victor’ distance measure tends to be better than van Rossum’s distance measure.

第1章 緒論 1
第2章 研究背景知識 3
第2.1節 VICTOR提出的度量方法 3
第2.2節 VAM ROSSUM提出的度量方法 6
第2.3節 改善VAM ROSSUM之度量方法 9
第2.4節 值譜分群法 9
第3章 實驗設計與方法 12
第3.1節 介紹原始資料 13
第3.2節 模擬棘波-棘波刪除 13
第3.3節 模擬棘波-棘波位移 14
第3.4節 JACCARD指標 17
第4章 實驗結果 19
第4.1節 棘波個數差異對於區別棘波序列之影響 20
第4.2節 模擬棘波刪除與位移對於區別棘波序列之影響 27
第5章 結論與未來展望 36
第5.1節 結論 36
第5.2節 未來展望 37
附錄 38
A. 公式推導 38
B. 實驗數據(二) 43
參考文獻 43

[1]F. R. K. Chung, "Spectral graph theory," Number 92 in CBMS Regional conference Series in Mathematics. American Mathematical Society, 1997.

[2]D. Greene, A. Tsymbal, and N. Bolshakova, "Ensemble clustering in medical diagnostics," 2004..

[3]Cai Xiao-yan, Dai Guan-zhong and Yang Li-bin, "Survey on Spectral Clustering Algorithms," Computer, p. 07, 2008.

[4]C. Houghton, "Studying spike trains using a van Rossum metric with a synapse-like filter," Journal of computational neuroscience, vol. 26, pp. 149-155, 2009.

[5]A. Kuhn, A. Aertsen, and S. Rotter, "Higher-order statistics of input ensembles and the response of simple model neurons," Neural Computation, vol. 15, pp. 67-101, 2003.

[6]A. Ng, M. Jordan, and Y. Weiss, "On spectral clustering: Analysis and an algorithm," pp. 849-856, 2001.

[7]A. R. C. Paiva, I. Park, and J. C. Principe, "A comparison of binless spike train measures," Neural computing & applications, vol. 19, pp. 405-419, 2010.

[8]A. R. C. Paiva, S. Rao, I. Park, and J. C. Principe, "Spectral clustering of synchronous spike trains," pp. 1831-1835, 2007.

[9]M. van Rossum, "A novel spike distance," Neural Computation, vol. 13, pp. 751-764, 2001.

[10]J. D. Victor and K. P. Purpura, "Metric-space analysis of spike trains: theory, algorithms and application," Network: computation in neural systems, vol. 8, pp. 127-164, 1997.