Semi-supervised ranking SVM-assisted IGA with hierarchical evaluations

2012 
Interactive genetic algorithms (IGAs), combining the human's intelligent evaluations with the evolutionary optimization mechanisms, have been sufficiently investigated due to their powerful performance in solving problems with aesthetic criteria. User fatigue arising from intelligent evaluations, however, greatly restricts IGAs to successfully optimize complicated problems. Lot of efforts have been devoted to effectively alleviating user fatigue and improving the exploration performance by designing friendly evaluation modes or surrogate models. A large population IGA with a ranking learning based surrogate model is presented here. A hierarchical evaluation mode reflecting the preference orderings of the user is first presented. From the perspective of directly maintaining the correct preference ordering, a ranking function is constructed to estimate the fitness of all individuals. An enhanced semi-supervised ranking SVM is developed to obtain the ranking function with high prediction accuracy, in which the method of selecting most informative unlabeled samples is highlighted by naturally combining the merits of IGAs. The ranking function is managed based on a defined metric of prediction accuracy. The proposed algorithm is applied to a fashion design system to empirically demonstrate its strength in alleviating user fatigue and improving exploration.
    • Correction
    • Source
    • Cite
    • Save
    • Machine Reading By IdeaReader
    27
    References
    0
    Citations
    NaN
    KQI
    []