Balancing samples’ contributions on GA learning
Refereed conference paper presented and published in conference proceedings


全文

其它資訊
摘要A main branch in Evolutionary Computation is learning a system directly from input/output samples without investigating internal behaviors of the system. Input/output samples captured from a real system are usually incomplete, biased and noisy. In order to evolve a precise system, the sample set should include a complete set of samples. Thus, a large number of samples should be used. Fitness functions being used in Evolutionary Algorithms usually based on the matched ratio of samples. Unfortunately, some of these samples may be exactly or semantically duplicated. These duplicated samples cannot be identified simply because we do not know the internal behavior of the system being evolved. This paper proposes a method to overcome this problem by using a dynamic fitness function that incorporates the contribution of each sample in the evolutionary process. Experiments on evolving Finite State Machines with Genetic Algorithms are presented to demonstrate the effect on improving the successful rate and convergent speed of the proposed method.
著者Leung K.S., Lee K.H., Cheang S.M.
會議名稱4th International Conference on Evolvable Systems: From Biology to Hardware, ICES 2001
會議開始日03.10.2001
會議完結日05.10.2001
會議地點Tokyo
會議國家/地區日本
出版年份2001
月份1
日期1
卷號2210
出版社Springer Verlag
出版地Germany
頁次256 - 266
國際標準書號354042671X
國際標準期刊號1611-3349
語言英式英語

上次更新時間 2022-14-01 於 23:52