Invited Talk: Multi-Component Nonnegative Matrix Factorization for Data Clustering

Dr.  Feng Tian

Professor, Bournemouth University, UK

  Dr Feng Tian is currently a professor in Bournemouth University, UK. With expertise on digital media, image processing and machine learning, Dr Tian has published over 100 papers or book chapters in peer-reviewed journals or international conferences, including IEEE Transactions on Visualization and Computer Graphics, ACM Transactions on Modelling and Computer Simulation, IEEE Transactions on Cybernetics, Visual Computer, Computer & Graphics, Multimedia Tools & Applications, International Joint Conference on Artificial Intelligence (IJCAI), Association for the Advancement of Artificial Intelligence (AAAI), Pacific Graphics (PG), IJCNN, CASA, CGI, etc. Before coming to the UK, Dr Tian worked as a post-doctoral fellow and assistant professor in Nanyang Technological University, Singapore. Dr Tian has also been awarded with research grants from Singapore National Research Foundation (Singapore), Royal Society (UK), British Art Council (UK), Horizon 2020 (EU), etc.

Abstract of Feng Tian's Talk

  A good data representation can typically reveal the latent structure of data and facilitate further processes such as clustering, classification and recognition. Nonnegative matrix factorization (NMF) as a fundamental approach for data representation has attracted great attentions. Despite its great performance, traditional NMF fails to explore the semantic information of multiple components as well as the diversity among them, which would be of great benefit to understand data comprehensively and in depth. In fact, real data are usually complex and contain various components. For example, face images have ex-pressions and genders. Each component mainly reflects one aspect of data and provides information others do not have. In this talk, I will present an approach on multi-component nonnegative matrix factorization (MCNMF). Instead of seeking only one representation of data, MCNMF learns multiple representations simultaneously, where each representation corresponds to a component. By integrating the multiple representations, a more comprehensive representation is then established.