Electric Engineering & Computer Sciences
Queen Mary, University of London
Roberts Building 508 (<map>)
Topic models in computer vision
Topic models are a category of machine learning methods that have now found wide application in supervised and unsupervised modelling of image, text and multi-modal data. In this talk I will give a non-technical overview of topic models and introduce how they can be applied to a variety of computer vision modelling problems. In particular, I will focus on a line of work in our lab that has applied topic models to analysing complex images and videos from a variety of angles. These include unsupervised scene understanding, abnormality and saliency, classification and annotation, detection and localisation, learning from sparse data, learning from weakly-supervised data, learning from noisy annotations, fusing multi-modal data, attribute learning and zero-shot learning, transfer learning and domain adaptation. If there is time and interest, I can also summarise some trade tricks about how to get good results with them in practice.
Page last modified on 08 feb 14 16:00