ICSA Talks - 09/05/19

Milos Nikolic - ICSA associate member

Title: Towards a Unifying Framework for In-Database Learning

Abstract: In this talk, I will first give a short overview of my research on building systems for real-time analytical processing. I will then focus on the topic of in-database machine learning over relational databases. I will demonstrate how one can cast a machine learning problem into a database problem by decomposing the learning task into a batch of SQL aggregates and by computing these aggregates over the input database. This approach benefits tremendously from structural properties of the relational data, factorized query evaluation, and query compilation. In practice, this translates to several orders of magnitude better performance than state-of-the-art data management and machine learning systems.

Bio: Milos Nikolic is a lecturer in the School of Informatics at the University of Edinburgh. Milos' research interests are in the areas of data management and stream processing systems, incremental computation, machine learning, and compilers. My work studies the incremental computation of complex analytical queries, such as SQL queries, linear algebra programs, and in-database machine learning tasks, in local and distributed streaming environments using novel approaches to query optimization and compilation. Before joining Edinburgh, I was a departmental lecturer in the Department of Computer Science at the University of Oxford. I received a Ph.D. in Computer Science from EPFL.

Tariq Elahi - ICSA associate member

Title: From Design to the Real World: Enabling Visibility into Sensitive Environments to Enhance Security and Privacy

Abstract: Privacy-enhancing technologies (PETs), such as anonymous communication networks (ACN), censorship resistance systems (CRS), and data privacy schemes (DPS), protect against mass surveillance, non-consensual online tracking, and the chilling of free speech.

However, these systems are often broken when the assumptions that their designs depend on do not hold in the real world. Assumption failures include adversaries with more power and reach than anticipated, client and network behaviour that does not conform to expectations, and the inherent meta-data leakage of the underlying networking technologies that these systems are built upon.

In this talk I will discuss my research exploring the arms race between adversary and the defender, using the Tor anonymous communication network as an example, and motivate for the need for empirical data extracted from real-world deployments. I will present our latest work in this direction where we make novel use of traffic analysis techniques, that have been typically seen as attacking privacy, as the means to measure sensitive phenomena in a privacy-preserving manner.

Bio: Tariq is a Lecturer in Security and the Internet of Things at the University of Edinburgh. He analyses, breaks, and enhances the security and privacy of systems attached to the Internet. His prior work centred around securing real-world anonymous communication systems against global adversaries. His current interests are in privacy-preserving data analytics and translating privacy and censorship resistance techniques to resource-constrained local-area networks.