According to the World Health Organization, recent years have seen a dramatic increase in the number of car accidents worldwide. In an attempt to ameliorate this situation, the automotive and telematics industry has tried to develop technology that can help the drivers make better and safer decisions. One approach is to develop systems that give feedback to the driver by means of a driving score, so that the driver can analyze his driving habits. By considering sensing platforms embedded into either vehicles or smartphones, this paper models the driving score calculation task as a regression problem. Accordingly, we propose novel scoring functions that are generated through a metaheuristic for automatic program induction called genetic programming (GP). In addition to our proposal, we evaluate six other computational methods, three of which are based on works reported in the literature. Results show that the functions generated by the GP clearly outperform all studied competitors. Moreover, the functions are simple enough to be transferred to an on-board vehicle computer or to the smartphone of the driver. Given the white-box nature of the proposed scoring solutions, these could be implemented in the aforementioned sensing platforms by interested parties to offer better insurance plans, as a tool for parents to monitor young drivers, or by users of public or private transportation system to evaluate and report their experience. Presented results indicate that the average velocity is the most important variable for driver scoring.

The interpretation of medical images is a fundamental process for the diagnosis and treatment of patients. This process contributes determining the causes of symptoms as well as monitoring the effects of any treatment. Although the generation of medical reports from images is a complex task, deep learning strategies have been integrated with models that allow this arduous task to be tackled, achieving promising results. This work aims to present a compilation of the most outstanding deep learning strategies focused on the automatic generation of medical radiology reports from X-Ray images. Papers based on DenseNet, ResNet and VGG architectures, in combination with Long Short-Term Memories (LSTMs) and attention models, are analyzed in terms of the pre-processing strategies, databases used, model adaptations, and metric results. All these important findings are summarized in this survey, highlighting those models that reported the highest performance.