Sbdc2018 master slidedeck-final

Editor's Notes

• #14: Beter uitleggen met symbolen
• #15: Stop symbol after start + Bias Free stamp
• #17: Current health research is unethical for several reasons, Cohorts are small and homogenous When clinical trial was successful, the treatment gets accepted and handed out to the population, However, this population is very different from the cohort sample And by not taking into account someone’s lifestyle and social economic status The results of clinical trial studies cannot be trusted Because omitted variables, like income, activity measures and social network bias the coefficients of statistical analysis and conclusions are likely to be wrong Unfortunately these social and economic data are hard to get by, and discovery of true, meaningful associations are hindered by a lack of access to data that are not easy to get at
• #19: In P8, we study access-restricted vertically partitioned data, where health data (from the maastricht study) and social economic data (from CBS) of the same individuals are stored at separate sites can be analyzed, simultaneously, without breaching privacy. Our use case is to uncover social and economic determinants of type II diabetes. To do so, we are creating a technically sound and secure infrastructure, using a Trusted Secure Environment to analyse these data. However, our future goal is to learn associations between these data sets, distributedly
• #44: -I'm going to tell you how you guys can improve the detection of bad guys amongsy a majority of good guys -This research is applied on the inspection process within the dutch foreign nationals employment law -We want to find the bad guys who illegaly hire foreign nationals -Limited resources->There are too many companies to inspect them all -inspection on a selection of the companies -Selecion is based on ranking that is the result of knowledge and experience of the domain experts -Based on this ranking the highest ranked companies are inspected -How can we make future inspection more effective by using the obtained results -What do the inspection results tell us about the remaining uninspected set
• #45: -We can exploit the domain knowledge hidden in the results by applying artificial intelligence -When we are training ai on only the inspection result we would need to compensate for the fact that only the highest ranked companies are inspected -However, majority of uninspected set are good guys -> why not use them as good guys? -Significant improvement achieved -But we can do even better: good guys in inspection are outliers in the sense that they look like bad guys -> discarding gives even better detection result -But we can do even better: inspected good guys look like inspected bad guys because they have been included for inspection. These are likely to be outliers, why not discard these inspected good guys. -Discarding known good guys from inspection results improves detection of the bad guys. -Thus we significantly improved bad guy detection by exploiting expert knowledge.
• #49: Rewarding industry with risks Production problems can occur every second in remote location (requires fast reaction)
• #60: Praatplaat voor de pitch (ongeveer 300 woorden)