Active learning based on a hybrid neural network modeller

  • Yonglong Wang

Student thesis: Doctoral Thesis


Various methods are investigated for selecting training data for the purpose of training neural networks. A new method called MIQR (Maximum Inter-Quartile Range) is proposed for effectively selecting a concise set of training data. In addition, the ensemble concept is introduced in this new method. Data selection is not unduly influenced by “outliers”, rather, it is principally dependent upon the “mainstream ” output of the ensemble networks. Encompassed in the new method is a very simple ancient Chinese philosophical idea, i.e. “the minority obeys the majority”. These techniques are nonparametric in the sense that several different neural networks comprise an ensemble or committee and co-operatively work together with each other to achieve a common goal. Because these are different neural networks (hybrid model), they can be complementary in the entire learning system, and therefore effectively enhance the entire learning system’s efficiency and accuracy. For learning, the neural networks attempt to actively select the most informative and important training data.

The methods described in this thesis pleasingly satisfy this need, and compare favourably with contending methods. Many experiments have been done to corroborate theoretical and empirical conjectures. The results are quite pleasing in that this new method is not only as “active learning” much better than “passive learning” both in data selection and in generalisation performance, but also outperforms other existing contending active learning methods. In particular, the results are very satisfying and interesting when the method is applied to discontinuous functions. Although the experiments are conducted with clean data selection, it should be easy to extend them to noisy data selection since the method developed is validated using unlabelled data. The algorithm developed for these methods has been rigorously tested, and proves to be highly autonomous and robust. The methods developed here are not restricted to use on neural networks. More generally, they can be applied to other scientific research and economic fields, even educational and sociological behaviour.
Date of AwardJun 1998
Original languageEnglish

Cite this