RESUMO
Fairness, through its many forms and definitions, has become an important issue facing the machine learning community. In this work, we consider how to incorporate group fairness constraints into kernel regression methods, applicable to Gaussian processes, support vector machines, neural network regression and decision tree regression. Further, we focus on examining the effect of incorporating these constraints in decision tree regression, with direct applications to random forests and boosted trees amongst other widespread popular inference techniques. We show that the order of complexity of memory and computation is preserved for such models and tightly binds the expected perturbations to the model in terms of the number of leaves of the trees. Importantly, the approach works on trained models and hence can be easily applied to models in current use and group labels are only required on training data.
RESUMO
BACKGROUND: The United Arab Emirates (UAE) is faced with a rapidly increasing burden of non-communicable diseases including obesity, diabetes, and cardiovascular disease. The UAE Healthy Future study is a prospective cohort designed to identify associations between risk factors and these diseases amongst Emiratis. The study will enroll 20,000 UAE nationals aged ≥18 years. Environmental and genetic risk factors will be characterized and participants will be followed for future disease events. As this was the first time a prospective cohort study was being planned in the UAE, a pilot study was conducted in 2015 with the primary aim of establishing the feasibility of conducting the study. Other objectives were to evaluate the implementation of the main study protocols, and to build adequate capacity to conduct advanced clinical laboratory analyses. METHODS: Seven hundred sixty nine UAE nationals aged ≥18 years were invited to participate voluntarily in the pilot study. Participants signed an informed consent, completed a detailed questionnaire, provided random blood, urine, and mouthwash samples and were assessed for a series of clinical measures. All specimens were transported to the New York University Abu Dhabi laboratories where samples were processed and analyzed for routine chemistry and hematology. Plasma, serum, and a small whole blood sample for DNA extraction were aliquoted and stored at -80 °C for future analyses. RESULTS: Overall, 517 Emirati men and women agreed to participate (68% response rate). Of the total participants, 495 (95.0%), 430 (82.2%), and 492 (94.4%), completed the questionnaire, physical measurements, and provided biological samples, respectively. CONCLUSIONS: The pilot study demonstrated the feasibility of recruitment and completion of the study protocols for the first large-scale cohort study designed to identify emerging risk factors for the major non-communicable diseases in the region.