Applying the Transfer Learning Techniques on Classification Model - A Brief Survey

The major conjecture in traditional machine learning process is both the training data and test data. The future test data must have same dissemination and it should contain same feature space. This identical conjecture may not be applicable to many real world applications. The identical conjecture might be infringed when a new data emerges from another new domain while the older domain has their own labelled data. One of the domains may contain classification task and ample training data may persist in another domain and also the last-mentioned data may exist in different feature space and different data distribution concept is followed. The method of labelling is more expensive and the process of transfer learning can be applied to become cost effective. The learning performance can be improved and the labelling efforts can be reduced by transfer learning concept. The Boosted Transfer Learning (TrAdaBoost) concept is established and the users are permitted to exploit only a few amounts of recently labelled data to hold the old data to produce the accurate classification model for the new data with high quality. The Boosted Transfer Learning is compared with inductive transfer learning, transductive transfer learning and unsupervised transfer learning to prove that TrAdaBoost effectively transfers the knowledge from old to new data.