"Since I'm personally very interested in research, it is really enjoyable for me to do a Ph.D. Not only can you study the topic you find fascinating, but you also get a decent scholarship and receive a highly valued degree after concluding your studies. I chose in particular to study at the computer science department in Paderborn due to matching research interests, as well as the department's and my supervisors' excellent reputation."– Peter Janacik, PACE PhD-student

Current student projects


Below you will find examples of short abstracts of the type of doctoral research currently being done by students at PACE.  We hope this will give you an idea of the type of project you could be doing if you choose to complete your PhD in Paderborn.

Computer Science

Name:Michael Allwright
Country:Australia
Studies:Swarm Intelligence

Social insects such as Ants, Wasps and Bees are capable of cooperating to build large scale and highly intricate nests. In these insect societies, there is no single agent or insect managing the construction, nor is there a blueprint in existence that defines where tunnels and chambers are built. The field of research, Swarm Intelligence, offers an explanation to how these individual insects achieve this task through simple rules and local interactions, and demonstrates how the global outcome of nest construction emerges from these simple rules and local interactions. 

Recent work in Swarm Intelligence has shown that is also possible to not only to model these insect societies, but to also reimplement them using robotic agents, this area of research amounts the concept of Swarm Robotics. 

In this research project, the principles of Swarm Intelligence and Swarm Robotics will be applied to the construction task to create an distributed autonomous robotic system that is capable of building some basic structures such as walls, towers and pyramids. To ensure that the system is robust, scalable and parallel, the principles of Swarm Intelligence, such as only working with simple rules and only allowing local interactions, will be strictly adhered to.

Name:Jahanzeb Anwer
Country:Pakistan
Studies:Computer Science

Radiation-induced errors in FPGAs are a great concern for its manufacturers and the semiconductor industry. A widely used technique to counter these errors is triple modular redundancy (TMR) which varies in its configuration depending on the placement of voters and the number of logic design stages. The higher the TMR reliability is required, the more is the area consumption, power dissipation and latency.

We believe that the FPGA should have the capability to decide at runtime, the level of TMR that fulfils its reliability requirements. For example, if certain module in a design is more critical, the circuit designer may use a higher level of redundancy for it while relying on lower levels for less critical modules. In short, the FPGA should be able to compute the trade-off among reliability level, area consumption, power dissipation and latency at runtime and dynamically reconfigures itself in the varying radiation-based environment.

We target the implementation of this concept with the operating system ReconOS which provides us the capability to program hardware threads of varying size at runtime that helps us in implementing the dynamic reliability concept. By mapping TMR copies and voters in different hardware threads, we can even counter thermally-induced errors in addition to radiation based ones.


 

Name:Christian Brenner
Country:Germany
Studies:Computer Science

In modern technical systems, software often realizes a major part of an increasingly complex functionality. Due to this growing complexity, it becomes harder to ensure correctness of the software. Especially in safety-critical areas, like transportation or medicine, failures in the software can lead to harmful accidents. Therefore, software engineers need support for developing software that does exactly what it is supposed to do.

For developing a correct system, engineers first need a way to specify the system's intended behavior. However, engineers usually cannot understand the complete system at once. For this reason, scenario-based specification techniques have been developed. Scenarios define how a part of the system is supposed to behave in one specific situation. By concentrating on one scenario at a time, the engineer is faced with less complexity at once. However, these scenarios still need to be turned into a correct software. If this is done manually by programming, the programmer needs to consider the interrelations of the various scenarios, which also is way too complex. Instead, to ensure that the system actually fulfills the specification, a correct implementation should be derived automatically.

The goal of my research project is to develop an approach for automatically deriving a distributed implementation model from a scenario-based specification of a real-time system. While the problem has already been solved for the centralized case, distributed systems pose additional challenges. In this kind of systems, no individual component has full information about the state of the complete system. For fulfilling the specification, the subsystems have to share information by communication. Therefore, I need to develop an algorithm to determine, which additional communication is needed. A further challenge is that this algorithm has to take into account real-time constraints of the system to make sure it reacts in time.

Name:Tobias Graf
Country:Germany
Studies:Computer Science

In recent years Monte-Carlo-Tree-Search has becoming a major algorithm for planning in domains, which are too complex to be solved deterministically. By randomly sampling and building a search tree, more and more problems can be tackled in a heuristic way. It's most prominent application is in the field of computer-go, but - as a general purpose algorithm for planning in MDPs - it is also applied to a variety of problems in intelligent systems including stochastic shortest path planning, feature selection, mixed integer programming, physics simulation and scheduling. One major problem of MCTS is to use the information gained by randomly sampling in the most efficient way. Adapting the way of sampling with the help of the information in the search-tree is a major challenge to achieve better efficiency, which is especially relevant for applying MCTS in computer-go or other domains which can be divided into several subproblems.

My research focuses on improving the quality of random sampling policies in MCTS. On the one hand, this includes improving the simulation policy by incorporating offline knowledge learned from experts. On the other hand, a combination of MCTS with classical reinforcement learning algorithms can adapt the simulation policy with the help of the experience gained in the search tree.

Name:Shouwei Li
Country:China
Studies:Computer Science

How to cope with computational intractability is a crucial problem in theoretical of computer science. Several methods have been developed in order to solve this problem, such as approximation algorithms, average-case analysis algorithms, randomised algorithms and heuristic methods. But all of these methods have their drawbacks. It is clearly that the direct way of attacking NP-hard problems is in providing deterministic algorithms which have exponential running times.

Currently, there is an increasing interest in faster exact solutions for NP-hard problems called parameterised complexity theory, whose leitmotif can be characterised by the words “not all forms of intractability are created equal”, is another proposal on how to cope with computational intractability in some cases. This method focus on classifying computational problems according to their inherent difficulty with respect to multiple parameters of the input and measures the complexity as a function in those parameters. This allows to classify NP-hard problems on a finer scale than in the classical setting, where the complexity of a problem is only measured by the number of bits in the input.

Name:Christine Markarian
Country:Lebanon
Studies:Computer Science

Unlike wired or cellular networks, wireless ad-hoc networks have no backbone infrastructure. Nodes communicate via peer-to-peer communications either through single hops or multihops. A connected dominating set (CDS) is commonly used as a virtual backbone for these networks. A CDS asks for a connected subset S of nodes in the graph such that each node in the graph is either in S or has a neighbor in it. A small CDS is desirable to reduce control overhead in the network. The minimum CDS problem is NP-hard in both general and Unit Disc Graphs.

A lot of research has been done on Unit Disc Graphs in which each node has the same transmission range. In practice, however, the transmission ranges of nodes need not be the same, due to differences in functionality, power control, or topology control. Therefore, a digraph or a Disc Graph (DG), in which nodes have different transmission ranges, would better model these differences.

Motivated by this, we aim to construct local algorithms for wireless ad-hoc networks modeled as digraphs and disc graphs for problems such as dominating set-based described above. On the other hand, we look at online algorithms for such problems in which we consider the dynamic behaviors of nodes in a wireless ad-hoc network

Name:Pavel Podlipyan
Country:Russia
Studies:Computer Science

Distributed computing is a field of computer science that studies computational complexity issues that emerge in systems consisting of autonomous entities interacting with each other in order to solve together a problem or execute a task. Unlike in studies related to concurrency in distributed systems, such as parallel computing, we focus our research on autonomy of entities and decentralization of the system. The distributed systems we are interested in do not contain any controllers external to the system, as well as there is no pre-defined controller inside of the system. Traditionally the entities have been assumed to be stationary and able communicate through some medium as in the web, peer-t-peer networks or other communication networks. However there is big class of systems where entities are also mobile and do not have explicit communication. The system of autonomous mobile robots is one of representatives of such class.

In our project the research goal is to investigate what abilities shall be granted to the group of autonomous robots and what algorithm shall be executed by each robot in order to accomplish in a distributed way a given task. Assuming weakest robots we want to find out is it even possible to accomplish particular task and if it is, then how fast robots can do it. Particularly we are interested how locality restrictions, namely the absence of global information influence the performance of the system of autonomous mobile robots.


 

Imprint | Webmaster | Recent changes: 10.07.2015