This thesis looks at how to characterize weaknesses in machine learning models that are used for detecting privacy-sensitive data in images with the help of crowdsourcing. Before we can come up with a method to achieve a goal, we first need to make clear what we consider privacy-sensitive data. We took the General Data Protection Regulation (GDPR) as a starting point, and performed a crowdsourcing task to see how workers interpret this regulation. Interpreting legal texts can be difficult, there is room for interpretation and the perception of a legal text can change over time. Therefore, we need to take the input of the crowd, next to our own input, to operationalize this regulation to use in this context. Next, we took a machine learning model for detecting privacy-sensitive data in images in order to retrieve saliency maps, which helps us with explaining the inner-working of the model. Subsequently, the saliency maps are inspected through a crowdsourcing task, with the established privacy definition, to find out the strengths and weaknesses. From the results, we see that crowd workers can be efficiently used to find the strengths and weaknesses of a machine learning model, while keeping the privacy definition in mind. Workers are able to consistently apply their views about privacy across different images, whilst also increasing the trust people have in the machine learning model. This shows us that we can use crowdsourcing efficiently in a fairly difficult context of privacy, and paves the way for a more sophisticated approach to privacy-sensitive elements in images, and even for contexts other than privacy.

Scientific advances in the field of artificial intelligence, and the ever increasing processing power of computers, have opened up opportunities to use artificial in- telligence for new industries and applications. Blockrise foresaw opportunities in using artificial intelligence for digital asset management. Blockrise’s founders and the development team formalised the problem and drew up a project pro- posal, which was accepted by the TU Delft Bachelor End Project coordinators and overseen by Prof. Dr. Jan Rellermeyer in the role of TU Coach.
The development team first reserved roughly two weeks to spend on research- ing the problem and possible solutions. A research proposal was formalised, in which details of the problem were explained and abstract solutions were given. During the next six weeks, the team started the concrete implementation of the solution. The team created a neural network to use as core functionality in the product. Supporting functionality was simultaneously developed to allow re- trieval and processing of necessary data. With enough useful data on hand, the neural network could be trained to make predictions based on an asset’s closing price, opening price, the highest price and the lowest price for each next minute, hour and day.
Extensions to the product were made in the form live-data processing, and validation and visualisation of predictions. Trading strategies were included to allow fully automated decision on placing market orders. The final stage of the development period was spent tweaking the neural network parameters in order to minimise prediction error.