TY - GEN
T1 - Fitness landscape-based parameter tuning method for evolutionary algorithms for computing unique input output sequences
AU - LI, Jinlong
AU - LU, Guanzhou
AU - YAO, Xin
PY - 2011
Y1 - 2011
N2 - Unique Input Output (UIO) sequences are used in conformance testing of Finite state machines (FSMs). Evolutionary algorithms (EAs) have recently been employed to search UIOs. However, the problem of tuning evolutionary algorithm parameters remains unsolved. In this paper, a number of features of fitness landscapes were computed to characterize the UIO instance, and a set of EA parameter settings were labeled with either 'good' or 'bad' for each UIO instance, and then a predictor mapping features of a UIO instance to 'good' EA parameter settings is trained. For a given UIO instance, we use this predictor to find good EA parameter settings, and the experimental results have shown that the correct rate of predicting 'good' EA parameters was greater than 93%. Although the experimental study in this paper was carried out on the UIO problem, the paper actually addresses a very important issue, i.e., a systematic and principled method of tuning parameters for search algorithms. This is the first time that a systematic and principled framework has been proposed in Search-Based Software Engineering for parameter tuning, by using machine learning techniques to learn good parameter values. © 2011 Springer-Verlag.
AB - Unique Input Output (UIO) sequences are used in conformance testing of Finite state machines (FSMs). Evolutionary algorithms (EAs) have recently been employed to search UIOs. However, the problem of tuning evolutionary algorithm parameters remains unsolved. In this paper, a number of features of fitness landscapes were computed to characterize the UIO instance, and a set of EA parameter settings were labeled with either 'good' or 'bad' for each UIO instance, and then a predictor mapping features of a UIO instance to 'good' EA parameter settings is trained. For a given UIO instance, we use this predictor to find good EA parameter settings, and the experimental results have shown that the correct rate of predicting 'good' EA parameters was greater than 93%. Although the experimental study in this paper was carried out on the UIO problem, the paper actually addresses a very important issue, i.e., a systematic and principled method of tuning parameters for search algorithms. This is the first time that a systematic and principled framework has been proposed in Search-Based Software Engineering for parameter tuning, by using machine learning techniques to learn good parameter values. © 2011 Springer-Verlag.
KW - Support Vector Machine
KW - Evolutionary Algorithm
KW - Problem Instance
KW - Finite State Machine
KW - Input String
UR - http://www.scopus.com/inward/record.url?scp=81855227252&partnerID=8YFLogxK
U2 - 10.1007/978-3-642-24958-7_53
DO - 10.1007/978-3-642-24958-7_53
M3 - Conference paper (refereed)
SN - 9783642249570
T3 - Lecture Notes in Computer Science
SP - 453
EP - 460
BT - Neural Information Processing : 18th International Conference, ICONIP 2011, Shanghai, China, November 13-17, 2011, Proceedings, Part II
A2 - LU, Bao-Liang
A2 - ZHANG, Liqing
A2 - KWOK, James
PB - Springer Berlin Heidelberg
T2 - 18th International Conference on Neural Information Processing, ICONIP 2011
Y2 - 13 November 2011 through 17 November 2011
ER -