Visualisation of Incomplete Data Using Class Information Constraints

We analyse how the training algorithm for the Generative Topographic Mapping (GTM) can be modified to use class information to improve results on incomplete data. The approach is based on an Expectation-Maximisation (EM) method which estimates the parameters of the mixture components and missing values at the same time; furthermore, if we know the class membership of each pattern, we can improve the generic algorithm by eliminating multi-modalities in the posterior distribution over the latent space centres. We evaluate the method on a toy problem and a realistic data set. The results show that our algorithm can help to construct informative visualisation plots, even when many of the training points are incomplete.