Data Driven Generation of Interactions for Feature Binding and Relaxation Labeling

We present a combination of unsupervised and supervised learning to generate a compatibilityin teraction for feature binding and labeling problems. We focus on the unsupervised data driven generation of prototypic basis interactions by means of clustering of proximityv ectors, which are computed from pairs of data in the training set. Subsequentlya supervised method recentlyin troduced in [9] is used to determine coefficients to form a linear combination of the basis functions, which then serves as interaction. As special labeling dynamic we use the competitive layer model, a recurrent neural network with linear threshold neurons, and show an application to cell segmentation.