Learning to Learn: Multi-Task learning, transfer learning and meta-learning

Multi-task learning, transfer learning, and meta-learning are three dimensions (sub-fields) of learning techniques, strategies and frameworks currently being investigated in machine learning communities. Multi-task learning research tries to investigate strategy to solve multiple tasks at the same time by exploring shared domain knowledge of two or more tasks. This also implies that a machine learning algorithm tries to induce a single model that may solve multiple tasks with reasonable accuracy. Multi-task learning closely links with the subfield transfer learning where a model induced on one problem domain is also used for solving another domain. For example, deep learning trained to identify cat can also be used for identifying animals in general. Meta-learning subfield studies the task of inducing a model on metadata such as general properties and distribution of a problem domain. This research call aims to investigate, explore and innovate in one or more of these research dimensions of machine learning research.




First degree in computer science, physics, engineering, and mathematics with 2:1 or above. MSc degree in the relevant subject areas is desired.