Ensemble Strategies for Semi-Supervised, Unsupervised and Transfer Learning

Traditionally there are two main paradigms in machine learning, supervised vs. unsupervised learning. A supervised learning algorithm uses teacher’s information (labelled examples) to train a learner while unlabelled data are automatically categorised by an unsupervised learning algorithm without using teacher’s information. In reality, however, labelled examples are often difficult, expensive, and/or time-consuming to obtain, which demands the efforts of experienced human annotators, while unlabelled data may be relatively easy to collect. Semi-supervised learning offers new techniques with the use of large amount of unlabelled data along with some labelled examples. In some situations, no labelled data are available so that one can only adopt the unsupervised learning paradigm for learning. Nevertheless, a common issue for both semi-supervised and unsupervised learning paradigms is how to exploit the information conveyed in unlabelled data. In a generic sense, the aforementioned learning problems may be naturally extended to transfer learning where other information sources can be explored to facilitate the current learning task in hand.

Ensemble learning studies machine learning algorithms and architectures that build collections of learners towards achieving better performance than an individual learner. This project is going to investigate typical ensemble learning methodologies, e.g., sequential and hierarchical combination of learning models, within the semi-supervised/unsupervised/transfer learning paradigms. The representation learning models that tend to tackle challenging real world problems that violate the standard yet conservative statistical assumptions made in the current machine learning algorithms. The main issues to be studied include theoretical/empirical investigation on novel ensemble representation learning framework including miscellaneous combination strategies in terms of generalization/stability and computational complexity, exploration/exploitation of unlabelled data or various information sources across different component learners and automatic model selection in the context of semi-supervised/unsupervised/transfer learning. In general, this project is suitable for one who is interested in fundamental research in machine learning while it is acceptable for one who already has a relevant application problem in mind and wishes to tackle their problems with an emerging technology such as ensemble learning. It is worth mentioning that this project description is generic and a specific project needs to be well-defined with a self-motivated students input.

In order to take this project, it is essential to have excellent mathematics and machine learning background knowledge as well as good programming skills. If you are interested in this project, please first visit my research student page: http://staff.cs.manchester.ac.uk/~kechen/ for the required materials and information prior to contacting me.