Feedforward networks use back propagation algorithm to train a multi - layer network . after training , the multi - layer network can fit the function in the data space very well 前向網(wǎng)絡(luò)利用反向傳播算法訓(xùn)練多層網(wǎng)絡(luò),使訓(xùn)練后的網(wǎng)絡(luò)較好地擬合樣本空間中各點的函數(shù)值。
The back - propagation algorithm also rests on the idea of gradient descent , and so the only change in the analysis of weight modification concerns the difference between t and y 反向傳播算法同樣來源于梯度降落原理,在權(quán)系數(shù)調(diào)整分析中的唯一不同是涉及到t ( p , n )與y ( p , n )的差分。
Then an improved interval propagation algorithm was brought up which solved the detection of multi - variable , value continuous constraints in aircraft cooperative design 基于這種樹形約束數(shù)據(jù)結(jié)構(gòu),文章提出了一種改進(jìn)的區(qū)間傳播算法,解決了飛行器協(xié)同設(shè)計中連續(xù)值域多元約束網(wǎng)絡(luò)中的沖突檢測。
The artificial neural networks ( ann ) with back propagation algorithms coupled with the sequential pseudo - uniform design ( spud ) was applied and demonstrated successfully to the modeling of the pmr system using limited but adequate experimental data 我們采用接續(xù)式擬均勻設(shè)計來安排實驗取得少量但充足的數(shù)據(jù)并以類神經(jīng)網(wǎng)路來建構(gòu)鈀膜反應(yīng)器之代表性模式。
To achieve this goal , this paper design a neural network with three layers in which the first layer play a classifier role and learn with the memory - based learning algorithm while the second and third layers learn with the error back - propagation algorithm 根據(jù)這一需要,本文建立了三層神經(jīng)網(wǎng)絡(luò),第一層起分類作用,采用基于記憶學(xué)習(xí)算法,第二、三層采用誤差反饋學(xué)習(xí)算法。
In the stage of training , nntcs applies labeled documents to ann for training , and the error back propagation algorithm ( bp ) is employed to adjust weights of the networks . after training , the final fixed weights are saved as knowledge of classification 在文本訓(xùn)練的時候,利用標(biāo)記好的訓(xùn)練文檔集進(jìn)行網(wǎng)絡(luò)訓(xùn)練,誤差反饋算法對網(wǎng)絡(luò)進(jìn)行權(quán)值調(diào)整,得到固定的權(quán)值作為分類知識存儲。
2 . after a brief introduction to belief propagation algorithm and a close research on the message flowing schedule , a new serial concatenated decoding method of ldpc codes based on the matrix decomposition and the two - way schedule is proposed in this thesis 在深入地研究了ldpc碼的基于因子圖模型的和積算法基礎(chǔ)之上,給出了一種雙向信息傳遞策略的實現(xiàn)方案,提出了一種ldpc碼的串行級聯(lián)譯碼算法。
In this paper , we improve the objective function of back propagation algorithm based on different financial actual situation . change the objective function into the expectation and the variance of error function to make its application more wide 本文首先基于bp算法應(yīng)用于金融實務(wù)領(lǐng)域的不同,對原bp算法的單一目標(biāo)函數(shù)進(jìn)行了改進(jìn),分別取其期望目標(biāo)和方差目標(biāo),進(jìn)行了bp算法的推廣,使其應(yīng)用范圍更廣。
In order to override the well - known limitation of back propagation algorithm , such as local grade problem , we suggest genetic algorithm , a global optimization algorithm , to optimize the weights set . the different parts of this model were modularized and combined as a prediction system 通過對固定網(wǎng)絡(luò)結(jié)構(gòu)的權(quán)系值進(jìn)行遺傳操作,優(yōu)化網(wǎng)絡(luò)的權(quán)系值組合,快速收斂到最優(yōu)權(quán)系值組合,進(jìn)而提高網(wǎng)絡(luò)的分析預(yù)測效率和能力。
It aims at gaining fast convergence with the same complexity as the conventional belief propagation algorithm . density evolution theory analysis and programming simulation results show that this new algorithm has fast convergence speed and good performance . 3 在詳細(xì)闡述了新算法的基本原理、譯碼結(jié)構(gòu)及實現(xiàn)步驟之后,利用理論分析和計算機(jī)仿真的方法,對串行級聯(lián)譯碼算法和傳統(tǒng)的置信傳播算法的性能做了比較,得出了一些有用的結(jié)論。