High performance deep learning libraries for biomedical applications

Deep learning approaches are a topic of growing interest since they can achieve high precision in machine learning tasks and may be useful in several scenarios, while high performance computing (HPC) is one of the driving factors for deep learning applications since they require massive computational power. One of these scenarios is related to biomedical context since the massive growth of data generated by several medical procedures. Deep learning techniques, applied on these data may be useful both for medical procedures and for further knowledge discovery in specific field (in example gene interaction related to some diseases). Therefore the importance to have a deep learning library tailored for these task is evident. This paper aims to discuss about some libraries specifically designed to provide convenient high performance computing oriented deep learning support to biomedical applications. We describe two libraries developed inside a European project, named the Deep Health Project, to support both deep learning basic operations and computer vision tasks, oriented to a distributed computing fashion and with some special features for managing biomedical data. In addition we highlight some differences and comparisons with popular environments like Keras and Tensorflow by describing a simple use case.