[4A3-J-1] Fundamental AI, theory: bio-inspired models of intelligence

[4A3-J-1-05] An extension method of semi-supervised boosting to multiclass classification

〇Yuta Sakai¹, Kazuki Yasui¹, Kenta Mikawa², Masayuki Goto¹ (1. Waseda University, 2. Shonan Institute of Technology)

Keywords:semi-supervised learning, boosting, multi-class classification, ensemble learning

Recent years, semi-supervised learning which classifies test data into correct category using not only training (labeled) data but a large number of unlabeled data has paid attention. However, in the semi-supervised learning setting, there is a problem that classification accuracy deteriorates because distribution of labeled data is biased. The SemiBoost is one of semi-supervised learning method to solve the problem. The SemiBoost is a binary classification method. However, this method can not be extended directly to multi-class classification. In this research, we propose the way to extend the SemiBoost for multi-class classification using the concept of Error Correcting Output Code (ECOC) method. To verify the effectiveness of our proposed method, we conduct simulation experiment using UCI machine learning repository.