A reoccurring task in the design and profiling of relational data is the discovery of hidden dependencies between attributes. Enumerating those dependencies is equivalent to the transversal hypergraph problem. We devise a backtracking algorithm for the enumeration of inclusion-wise minimal hitting sets. It achieves polynomial delay on hypergraphs for which the size of the largest minimal transversal is bounded. The algorithm solves the extension problem for minimal hitting sets as a subroutine. Although this problem is NP-complete and W[3]-complete in general, we show that a careful implementation of the extension oracle can help avoiding the worst case on hypergraphs arising in the profiling of real-world databases, leading to significant speed-ups in practice.

This is joint work with Thomas Bläsius, Tobias Friedrich, Julius Lischeid, and Kitty Meeks.

A time series (TS) is a collection of values sequentially ordered in time. TS emerge in many scientific and commercial applications. One driving force behind their rising importance is the sharply increasing usage of sensors for automatic and high resolution monitoring in domains such as smart homes, machine surveillance, land cover mapping using satellite images, smart grids, weather observations, wind energy forecasting, mobility tracking, etc.

The main purpose of TS analytics is to extract meaningful knowledge based on the shape of the TS data, which implies exploiting temporal order. Though humans have some natural capacity to solve the identification of perceptually similar but not mathematically identical TS, this remains a complex problem for computers.

In this talk, I will provide an overview of the techniques applied for TS analytics, where approaches can broadly be divided into Whole Series, Shapelets, Dictionary (Bag-of-Patterns), Ensembles, and Deep-Learning. With a focus on TS classification, I will address how state-of-the-art approaches represent the fundamental shapes of a TS and how these features are used. The talk ends with an experimental evaluation of the state-of-the-art classification approaches.

Many real-world networks share common properties like power law degree distribution, small diameters, and large clustering coefficients. Graph models used in research try to mimic real-world networks by providing similar properties. Hyperbolic random graphs in particular have recently gained popularity in network research because they implement all mentioned properties. However, generators for hyperbolic random graphs only support a limited number of dimensions for the underlying geometry, resulting in tight links between average node degree and clustering coefficients. A generalization of hyperbolic random graphs are geometric inhomogeneous random graphs (GIRGs), pro- posed by Bringmann et al. They also present a sampling algorithm to generate GIRGs that has an expected running time of O(n) that supports an arbitrary number of dimensions in the underlying geometry.

In this Thesis, we explain GIRGs and the sampling algorithm as proposed by Bringmann et al. We present the first implementation of a graph generator with underlying geometry that supports an an arbitrary numbers of dimensions. We suggest some enhancements to the sampling algorithm proposed by Bringmann et al. to boost the run-time of the generator by 20%. We also theoretically analyze configurations for the sampling algorithm to give the user of our tool more direct control of the output. Finally, we use our implementation to empirically validate theoretical findings on GIRGs.

Networks are a fundamental part of our society. The recent development of devices shows that more and more networks are experiencing a shift from wire-lined to wireless ad-hoc networks. Such networks exist in numerous forms and are from a multitude of different domains. By using the tools and explanatory power of Game Theory, we explore the creation and properties of such networks.

In this work we look at Wireless Network Creation Games, where selfish agents build an ad-hoc network. In contrast to most current research on this topic, which only tries to minimize the energy consumption of an agent, we also analyse the quality of an agent’s connection by creating a trade-off between energy consumption and experienced service quality. Because of that we define multiple games by additionally imposing a parameter similar to the hop distance of internet networking, which is used to measure the real-time performance and stability of connections in a network. In these games we study Nash Equilibria, as well as the resulting dynamics created by the behaviour of selfish agents. We also provide an approximation algorithm that efficiently converges and provide empirical evidence to show that the resulting states of the Wireless Network Creation Game are close to optimal. Furthermore, we analyse the social welfare of these networks to illustrate how the selfish behaviour of agents influences the well-being of society.

The very general problem of determining the quality of a given partial solution occurs basically in every algorithmic approach which computes solutions in some sense gradually. Pruning search-trees, proving approximation guarantees or the efficiency of enumeration strategies usually requires a suitable way to decide if a partial solution is a reasonable candidate to pursue. Consider for example the classical concept of minimal dominating sets for graphs. The task of finding a maximum cardinality minimal dominating set (or an approximation of it) as well as enumerating all minimal dominating sets naturally leads to solving the following extension problem: Given a graph G=(V,E) and a subset P of V, does there exists a minimal dominating set S for G which contains P.
In an attempt to study the nature of such extension tasks, we propose a general, partial-order based framework to express a broad class of what we refer to as extension problems. In essence, we consider optimisation problems in NPO with an additionally specified set of presolutions (including the solutions) and a partial order on those. We consider a number of specific problems which can be expressed in this framework. Possibly contradicting intuition, these problems tend to be NP-hard, even for problems where the underlying optimisation problem can be solved in polynomial time. This raises the question of how fixing a presolution causes this increase in difficulty. In this regard, we study the parameterised complexity of extension problems with respect to parameters related to the presolution. We further discuss relaxation of the extension constraint asking only for a solution S which preserves as much of P as possible. These considerations yield some insight into the various difficult aspects of extension problems.

Network Creation Games are a well-known approach for explaining and analyzing the structure, quality and dynamics of real-world networks like the Internet and other infrastructure networks which evolved via the interaction of selfish agents without a central authority. In these games selfish agents which correspond to nodes in a network strategically buy incident edges to improve their centrality. However, past research on these games has only considered the creation of networks with unit-weight edges. In practice, e.g. when constructing a fiber-optic network, the choice of which nodes to connect and also the induced price for a link crucially depends on the distance between the involved nodes and such settings can be modeled via edge-weighted graphs.

We incorporate arbitrary edge weights by generalizing the well-known model by Fabrikant et al. to edge-weighted host graphs and focus on the geometric setting where the weights are induced by the distances in some metric space. In stark contrast to the state-of-the-art for the unit-weight version, where the Price of Anarchy is conjectured to be constant and where resolving this is a major open problem, we prove a tight non-constant bound on the Price of Anarchy for the metric version and a slightly weaker upper bound for the non-metric case. Moreover, we analyze the existence of equilibria, the computational hardness and the game dynamics for several natural metrics. The model we propose can be seen as the game-theoretic analogue of a variant of the classical Network Design Problem. Thus, low-cost equilibria of our game correspond to decentralized and stable approximations of the optimum network design.

When I was a child, I wanted to program video games. In fact, this was one of the reasons why I decided to study computer science instead of mathematics. Due to my early ambitions, I regularly came into contact with C++, as it was arguably the biggest player out there. However, it was always a double-edged sword to me. On the one hand, I admired it for all of its power and features. On the other hand, I always had problems with anything I wanted to do, and I always felt insecure about what I was doing. In the end, I thought that I was too dumb to grasp the greatness of C++. Nonetheless, I always returned, hoping that I had matured and was now worthy of wielding the power that came with the best programming language on earth.

When I started my bachelor’s, I was introduced to Java. This made me question my world view multiple times. I actually learned how to program?! Apparently, doing a bachelor in computer science was more hand-holding than I thought. Also, programming could be quite easy?! Everyone who has worked with Java (or Python) probably knows how convenient it is to make progress. For most of the time, things work pretty well. And if they don’t, one usually knows where to start looking for errors.

Despite my good experience with Java, I longed for more. I was not very happy that everything in Java needed to be a class, and the execution speed of C++ was still better. Last, generic programming in Java is poorly implemented. I recall how my Java lecturer (who worked on heavily improving the Java VM) told us how much of a mess generic programming in Java was and how to use certain hacks in order to make it bearable. This was the point when I lost interest in Java. I saw a clear barrier built into the language that I could not surpass. In contrast to that, C++ was still out there. Waiting with all of its cool features to be explored and used. Limitless possibilities! Thus, I went back to looking at C++.

By now, I had learned many different programming concepts. I surely should be able to easily handle C++ now. But, boy, was I wrong! I basically struggled with the same things I had always struggled. Some things maybe made more sense. However, the way to go was always this long, winding road with tons of boulders placed brilliantly around you. Have fun escaping that! Nonetheless, people have used C++ for ages (relatively speaking). This was probably the price to pay for efficiency. Hence, I sat down and forced myself, again, through C++, this time trying really hard to pay attention to all of the details [1]. Then everything changed!

I was talking to a professor of one of the theory groups. He told me that a PhD student of his was planning to give a lecture on library design and wanted to present a new programming language that his group started using. Of course, I was intrigued and attended the class. The programming language used was called D – a lame pun on C++. I was not amazed. However, I got hooked very quickly with every new feature I was taught. D was very similar to Java, but it did not force you down the OOP road. In addition to that, D had learned from C++ and only tried to rely on backward compatibility with C if it was necessary. Last, it was designed with features in mind that could be considered ugly hacks in C++. D just hit all the right notes for me. It was C++ done right!

In my talk, I want to share my thoughts on D and give a broad overview over the language. I do not want to convince you that D is the best language out there (albeit I may be bashing C++ from time to time). My intention is to present you with an alternative you may want to consider in the future. I personally think that D has the potential of replacing C++ and Python for certain use cases.

While I am fully aware of the downsides of D, I will be focusing on the positives. However, you are motivated to challenge me anytime during the talk and point out bad design choices or questionable decisions. I would be happy to have some controversy going on. But be aware that I am not a programmer. I think that I have a rough idea of how idiomatic D looks like. However, I cannot claim the same for me about C++. When I looked at C++, smart pointers weren’t a thing and include guards asked for the most creative names imaginable. I would like to hear how things have changed and how C++ might be able to compete with D for some of the features I present. I am looking forward to your input.

[1] For example, I debated with myself for hours where to put the asterisk for pointers – a problem that probably many C++ programmers can relate to. I really wanted to put it directly after the type. Bjarne Stroustrup discusses this issue and says that both choices are fine (http://www.stroustrup.com/bs_faq2.html#whitespace), while he prefers his asterisks next to the type. However, looking at more complex examples (I am a theoretician after all) reveals that the only correct answer is that the asterisk should go immediately in front of the name, which I consider tremendously ugly. This is the only interpretation that satisfyingly explains how to declare more complex pointers, and it gets quite evident when looking at function pointers in namespaces. Thus, while one could argue that the original idea stems from C, which is thus to blame, namespaces are a C++ feature. Hence, Bjarne sort of overlooked this little detail. Admittedly, one is encouraged to not use the language way of declaring complex pointers but hacks instead. However, this only shows how certain features are misdesigned but will not be changed due to backward compatibility. Sorry for the rant.