Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/72277
標題: 多元雙向聯想記憶類神經網路之探討
作者: 冷明河
鄧洪聲
關鍵字: 類神經網路;雙向聯想記憶網路;多元雙向聯想記憶網路;鬆弛法
出版社: 國立中興大學工學院;Airiti Press Inc.
Project: 興大工程學報, Volume 10, Issue 1, Page(s) 59-82.
摘要: 
The Multi-BAM associative neural network is proposed in this paper, which is an extension of the Bidirectional Associative Memory (BAM). Instead of only two units, as in the BAM, the Multi-BAM consists of many units. It operates in an iterative way like that of the relaxation process. The energy of the network is also defined, and is proven to converge during the iterative process of the network. Based on the property that the energy of the network will reach its local minimum when all the units in the network are correct, we have derived the sufficient conditions that the network must meet in order that the network, when errors exist in its units, will be able to reach the correct recalls during iteration. Some experimental simulations have been done to investigate the network's performance under various error conditions. These experiments have also compared the effects of the network sizes and positions of the erroneous units in the network on the network performance.

在本篇論文中,我們提出了多元雙向聯想記憶綱路(Multi-BAM)的架構,它是由雙向聯想記憶網路擴大為具有多個單元的網路,其運作方式是與鬆弛法類似的疊代過程。我們也定義了網路的能量,並且證明了在疊代的過程中,Multi-BAM網路最後會收斂的結論。我們並利用網路總能量在所有單元完全正確時會達到能量局部最小值的特性,推導出當綱路中有單元發生錯誤時,網路經過疊代後可達到正確聯想時所必須滿足的充分條件。我們也用了幾個實驗模擬的例子來探討網路單元在發生了各種錯誤的情況下,綱路聯想的結果,也比較了綱路的大小及錯誤單元在網路中的位置對聯想結果之影響。
URI: http://hdl.handle.net/11455/72277
ISSN: 1017-4397
Appears in Collections:第10卷 第1期
工學院

Show full item record
 

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.