一個新的神經網路學習法─蟻窩演算法

Design of a New Neural Network Using an Ant Colony Optimization Technique

李正彬(Jeng-Bin Li)

指導教授 ： 鍾雲恭

元智大學/工程學院/工業工程與管理學系/碩士(2002年)

摘要

研究者與從業人員已經應用一些以生物導向技術的綜合方法於最佳化設計之問題上，諸如：旅行銷貨問題、自動控制、交通途徑管理與圖形辨識等。蟻窩最佳化（ant colony optimization , ACO）模仿真實世界裡螞蟻由食物來源處以最短的途徑搬運食物回窩巢中的行為，稱為生物演算法。類神經網路（artificial neural networks, ANNs）也是另一種生物演算法。本文發展一個以ACO訓練過程來獲得最佳連結權重值的新前饋式類神經網路。當螞蟻行經網路的連結線便殘留下費洛蒙濃度，此為主要的權重值決定因素。因此，當最終的權重值決定時，象徵由蟻窩式神經網路（ant-based network, ABN）所獲得的全連接式神經網路長度為最短。ABN是一個新的但自然的神經網路範例，所以本文以已發表過的三個時間序列模式來驗證、比較其它兩種典型的方法─傳統的倒傳遞神經網路（back-propagation neural network, BPN）與標準的統計迴歸技術。意外地發現，就統計上複判定係數與均方誤差來說ABN具有優越的預測能力。

關鍵字

蟻窩最佳化 ； 類神經網路 ； 蟻窩式神經網路 ； 倒傳遞神經網路

並列摘要

Researchers and practitioners have applied several biologically oriented techniques to be a comprehensive approach for optimal design problems, such as traveling salesman problems, robot control, communication routing management and pattern recognition. Ant colony optimization (ACO), which mimics the real behavior of forming the shortest path of food transportation from a food source and an ant nest in a life of real ants, is one of such the bio-computation techniques. Also, artificial neural networks (ANNs) is another bio-computation technique. This thesis develops a new feed-forwards ANN based on ACO training procedure to obtain the optimal connection weights in the ANN. The concentration of pheromone laid by the mimic ants moving on the ANN connection path is the key factor of the weights determination. This is the metaphor that the finalized determined weights make the total length among full-connected neurons of the proposed ant-based network (ABN) be the shortest. The ABN is a brand new but crude ANN paradigm that has been used to model three time series systems taken from the published papers, and the prediction abilities of these three known time series models are demonstrated by comparing them to those of other classical two methods, a traditional back-propagation neural network and a standard statistical regression technique. Surprisingly, the superiority of the ABN predictabilities in terms of both the statistical determination coefficient and the means square error is beyond the expectation.

並列關鍵字

Ant colony optimization ； Artificial neural networks ； Ant-based network ； Back-propagation neural network

參考文獻

Albus J. S., (1975) “Data Storage in the Cerebellar Model Articulation Controller (CMAC),” ASME Journal of Dynamic Systems, Measurement, and Control, Vol. 97, pp.228 -233.

Araüjo, F., B. Ribeiro and L. Rodrigues, (2001) “A Neural Network for Shortest Path Computation,” IEEE Transactions on Neural Network, Vol. 12, No. 5, pp.1067 -1073.

Cibas, T., F. F. Soulie, P. Gallinari and S. Raudys, (1996) “Variable Selection with Networks,” Neurocomputing, Vol. 12, pp.223 -248.

Dorigo, M. and L. M. Gambardella, (1997) “Ant Colony System: A Cooperative Learning Approach to the Traveling Salesman Problem,” IEEE Transaction on Evolutionary Computation, Vol. 1, No. 1, pp.65 -70.

Gray, S. M., J. Huang and C. J. Spanos, (1991) “Statistical Experimental Design in Plasma Etch Modeling,” IEEE Transactions on Semiconductor Manufacturing, Vol. 4, No. 2, pp.83 -98.