查詢結果分析
相關文獻
- 以實驗設計方法對倒傳遞網路學習參數進行最佳化的研究:於藥物動力學的應用為例
- 選擇性血清素再吸收抑制劑
- Fluvoxamine對Clozapine藥物動力學之影響
- 藥物動力學在臨床上之應用:藥物動力學服務
- 碳酸酐酶抑制劑6-Hydroxyethoxy-2-benzothiazole Sulfonamide靜脈注射後之兔眼藥物動力學
- 碳酸酐酶抑制劑6-Hydroxyethoxyzolamide局部點眼後之兔眼藥物動力學
- 新抗癲癇藥
- Pharmacokinetics and Brain Regional Distribution of caffeine in Rats
- 製程間具相關數據之趨勢模型偵測探討
- 口服Ketorolac藥物動力學特色
頁籤選單縮合
題 名 | 以實驗設計方法對倒傳遞網路學習參數進行最佳化的研究:於藥物動力學的應用為例=Experimental Design for Searching the Optimal Backpropagation Neural Network Configuration: Aplication to Pharmacokinetics as an Example |
---|---|
作 者 | 葉若春; 鄭春生; 卜玉枝; | 書刊名 | 管理與系統 |
卷 期 | 6:1 1999.01[民88.01] |
頁 次 | 頁65-91 |
分類號 | 418.1 |
關鍵詞 | 倒傳遞網路; 藥物動力學; 部份因子實驗設計; 網路學習參數; Backpropagation neural network; Population pharmacokinetics; Fractional factorial design; Design parameters; |
語 文 | 中文(Chinese) |
中文摘要 | 對於倒傳遞網在藥物學上的應用,至今仍缺乏系統性的研究方法,來探究其最適化網路參數的設計。針對此問題本文提出一個以解析度為IV之28-3部分因子實驗設計,對網路架構內的八個設計因子(A~H)進行研究。由常態機率分布圖的結果指出:主要因均不顯著,僅有兩個二因子交互作用(DH、AG+BD)具有顯著性。由於二因子交互作用交絡在一起,因此必須進行鏡射試驗使其相互分離。在結合主要試驗與鏡射試驗之後,將擁有足夠的情報以估算所有的主要因與二因子交互作用;此時方可建立合適的模式去設計確認試驗。在確認試驗的實測值與預估值極為相近,且滿足殘差的三大基本假設時,該實驗的適切性即可獲得驗證。因此本研究顯示:倒傳遞網路在人口藥物動力學上的應用,以部份因子實驗設計進行最佳化,確實為一可行且有效的分析工具。 |
英文摘要 | Currently there is lack of a systemic approach to select the optimal design parameters for pharmaceutical applications when the backpropagation neural networks are used in the field. In order to deal with this problem, a resolution IV 28-3 fractional factorial design is employed to investigate the effects of 8 design factors (A~H) on the neural network configuration. A result of the normal probability plot reveals that main effects are not significant, two largest effects are two-factor interactions (DH、Ag+BD). Due to confounding among the two-factor interactions, the principal fraction allows a sequential investigation to separate the two-factor interactions by foldovering the design. The combined design gives sufficient information to estimate all of the main effects and two-factor interactions, then we have developed the fitted model for designing the confirmation experiments. If the estimated values in the confirmation experiments are close to the expected intervals and the assumptions of the model are ascertained, then we can conclude that the combined design is fundamentally adequate. The result drawn from this research shows that the fractional factorial design provides an efficient means to optimize the design parameters of a backpropagation neural network applied in population pharmacokinetics. |
本系統中英文摘要資訊取自各篇刊載內容。