Speculative reordering based on neural networks for a latency-optimized privacy protection in complex event processing

Abstract

Nowadays, many devices are connected to the Internet. They are part of IoT applications, that offer us new services to improve our daily lives. A state-of-the-art paradigm to transform the raw data collected by these devices into meaningful information is Complex Event Processing (CEP). However, CEP systems often have access to privacy-sensitive information about the users, that they don’t want to expose. If an application is not able to conceal this information, it significantly reduces its acceptance. Access control is one technique for such privacy protection. Most access control mechanisms protect privacy only at the level of single attributes of data or events. However, sensitive information is often revealed via patterns of events. Palanisamy et al. [PDTR18] proposed such a pattern-level access control component, as part of a CEP application, that conceals private patterns by reordering events. This component achieves high Quality of Service (QoS) but has the downside that it incurs high latency when the window size is significantly large. The reason is that it processes the event stream window by window and can therefore only start reordering when all events of the current window are known. In this thesis, we extend the reordering approach to a speculative reordering strategy, that can already reorder before all events of the window are available. The evaluation results show that this can drastically reduce latency and also has other advantages.