Probabilistic and Bayesian Modelling I

There is a dramatic growth in the availability of complex data from a wide range of different applications. The challenge of the data analyzer is to extract knowledge from the raw data by identifying the useful patterns and structures that underlie it. This module introduces adaptive and probabilistic approaches to modeling such complex data. We first consider finding structure in high-dimensional data. The kernel methods approach to identifying non-linear patterns in introduced while addressing the issues of statistical reliability of inferences made from limited data. Subspace identification is considered and correlations across different data modalities are shown to provide a useful approach to eliciting semantic representations. The final section of the course will introduce learning probabilistic models, (e.g. in biological sequence data), fusing prior knowledge and data, complex and approximate inference.