Continuous and Discrete Deep Classifiers for Data Integration

Abstract

Data representation in a lower dimension is needed in applications, where information comes from multiple high dimensional sources. A final compact model has to be interpreted by human experts, and interpretation of a classifier whose weights are discrete is much more straightforward. In this contribution, we propose a novel approach, called Deep Kernel Dimensionality Reduction which is designed for learning layers of new compact data representations simultaneously. We show by experiments on standard and on real large-scale biomedical data sets that the proposed method embeds data in a new compact meaningful representation, and leads to a lower classification error compared to the state-of-the-art methods. We also consider some state-of-the art deep learners and their corresponding discrete classifiers. We illustrate by our experiments that although purely discrete models do not always perform better than real-valued classifiers, the trade-off between the model accuracy and the interpretability is quite reasonable.