Learning Co-embedded Spaces from Relations for Heterogeneous Objects
- Speaker: Dr Tingting Mu (University of Liverpool)
- Host: Sophia Ananiadou
- 11th March 2015 at 14:00 in Kilburn L.T. 1.4
The modelling of information as data points in multi-dimensional spaces is the fundamental stage in data analysis, reasoning, prediction and detection tasks. Particularly, learning a common low-dimensional space where heterogeneous objects can be treated homogeneously offers the opportunity of adapting a large number of established machine learning tools to solve challenging heterogeneous data analysis problems, in addition to denoising and redundancy reduction. In this talk, I will present our recent research on learning subspaces to encode numerical relations between heterogeneous objects. Application examples in text mining will be demonstrated, e.g., simultaneous and collaborative analysis of articles, phrases and semantic topics, and their visualisation.