• 作者： 黃仲陵
• 作者服務機構： 國立清華大學電機工程研究所
• 中文摘要： 本篇報告提供一種影像分類的新方法，非線性類比式神經網路在這方面有非常不錯的效果。我們的目的就是如何由原始影像得到一個最佳的分類結果。最佳化是依據神經網路的類比輸出所構成的能量函數獲得。神經網路由眾多的神經元構成，而神經元可想像成是一個具有曲狀－單調（sigmoid-monotonic）函數的放大器。連接神經元之間的神經鍵可視為是前一個神經元的輸出到下一個神經元的輸入間的電傳線，而流入神經元的總電流即為所有有連接前一級神經元輸出的總和。我們已將所希望的最佳化和一些限制條件用式子表示出來。
• 英文摘要： This paper develops a new method for pixel classification of an intensity image. A neural network ofnon-linear analog neurons has been shown to be extremely effective. This problem is considered as anoptimally classification of an image based on its original activation. Optimization is defined in terms ofenergy which is a function of neurons whose output values may vary continuously. The neurons aremodelled as amplifiers which have sigmoid monotonic input-output relations. A synapse between twoneurons is defined by a conductance which connects the output of a neuron to the input of anotherneuron. The net input current to any neuron is the sum of the currents flowing through the set of resistorsconnecting its input to the outputs of the other neurons. We have formulated the problems in terms ofdesired optima, subject to certain constraints.
• 中文關鍵字： pixel classification; hopfield model; neuron; sigmoid monotonic function
• 英文關鍵字： --