Final Report Summary - INFPROBMOD (Approximate Inference in Probabilistic Models)
In this project we have analysed and developed probabilistic models and approximate inference techniques for complex real-world applications. In particular, we have considered two important problems in robotics and bioinformatics: the creation of movement libraries for robot imitation learning, and the incorporation of family structure into genetic data analysis. There were two major achievements of the project:
- Introduction of an efficient method, requiring only limited human intervention, for segmenting time-series such as human movement data into basic actions. This represents a relevant contribution towards the goal of creating basic movement libraries, a problem that is receiving increasing attention in the Robotics community.
- Extension of a popular genetic data model for genotype imputation and haplotype estimation so as to account for mother-father-child relationships. This extension gives good performance, can be used by experimenters with little knowledge of modelling, and is computational efficient. The underlying approach can also be applied to more complex pedigrees. This work opens the way towards the more general use of deterministic approximate inference techniques for analysing genetic data from related individuals - a type of data that is becoming increasingly common, especially in the studies of complex diseases.
- Introduction of an efficient method, requiring only limited human intervention, for segmenting time-series such as human movement data into basic actions. This represents a relevant contribution towards the goal of creating basic movement libraries, a problem that is receiving increasing attention in the Robotics community.
- Extension of a popular genetic data model for genotype imputation and haplotype estimation so as to account for mother-father-child relationships. This extension gives good performance, can be used by experimenters with little knowledge of modelling, and is computational efficient. The underlying approach can also be applied to more complex pedigrees. This work opens the way towards the more general use of deterministic approximate inference techniques for analysing genetic data from related individuals - a type of data that is becoming increasingly common, especially in the studies of complex diseases.