QLTL: a Simple yet Efficient Algorithm for Semi-Supervised Transfer Learning

• Published in: 10th International Conference on Pattern Recognition Systems (ICPRS-2019) pp. 6 (30 - 35
• Main Authors: Muller, B, Lengelle, R
• Format: Conference Proceeding
• Language: English
• Published: Stevenage, UK IET 2019; The Institution of Engineering & Technology
• Subjects: Algorithms; Classification; Computer vision and image processing techniques; Document processing techniques; Exact solutions; Image recognition; Knowledge engineering techniques; Machine learning; Optimisation techniques; Other topics in statistics; Semi-supervised learning; Training; training set; text analysis; Quadratic Loss Transfer Learning; image classification; semisupervised learning; probability; similar-but-different underlying distribution; document image processing; convex programming; machine learning techniques; text-document classification task; semisupervised Transfer Learning; centred kernel polarisation criterion; QLTL; available training data; learning (artificial intelligence); density probability function shift
• ISBN: 1839531088; 9781839531088
• DOI: 10.1049/cp.2019.0244

Most machine learning techniques rely on the assumption that training and target data share a similar underlying distribution. When this assumption is violated, they usually fail to generalise; this is one of the situations tackled by transfer learning: achieving good classification performances on different-butrelated datasets. In this paper, we consider the specific case where the task is unique, and where the training set(s) and the target set share a similar-but-different underlying distribution. Our method, QLTL: Quadratic Loss Transfer Learning, constitutes semi-supervised learning: we train a set of classifiers on the available training data in order to input knowledge, and we use a centred kernel polarisation criterion as a way to correct the density probability function shift between training and target data. Our method results in a convex problem, leading to an analytic solution. We show encouraging results on a toy example with covariate shift, and good performances on a text-document classification task, relatively to recent algorithms.