Every year, ICSI hosts postdoctoral fellows from Germany who are funded by the German Academic Exchange Program. In September, we welcomed the first four postdocs to arrive in this year's program. They will be working on a variety of topics from theoretical algorithms to computer vision.

Michael Elberfeld started his PhD at the University of Lübeck by designing algorithms for computational problems that arise in molecular biology. Later he worked on the complexity theory of space-efficient and parallel computations. While time-efficient algorithms are fast, they often need huge amounts of memory for storing and manipulating data structures. Space-efficient algorithms trade time for space to save memory. In his PhD thesis he was interested in what problems can be solved using a logarithmic amount of space. While at ICSI, he's interested in how to find paths in networks using few space, a research agenda related to better understanding the difference between deterministic and nondeterministic space-bounded computations.

Christof Leng received his doctorate from the Technical University of Darmstadt a few weeks ago. For his thesis, he worked on BubbleStorm, a search system for peer-to-peer networks. BubbleStorm is a rendezvous search method, which allows users to search for data by using complex queries, like keyword search or even regular expressions. By contrast, distributed hash tables require that exact keys of values be known. In the simplest example of a rendezvous search, the nodes of a network are arranged in a grid and a piece of data is replicated on all points in a column. When a query is sent, it goes to all points along a row. The grid approach is not flexible enough to cope with large scale peer-to-peer networks, particularly as participants in the network come and go with high frequency. BubbleStorm deals with these complications by using stochastic algorithms. Christof identified different ways of replicating and updating data in this stochastic environment and used BubbleStorm for building an online game, a wiki, and a file-sharing program.
At ICSI, he will be working on methods of dynamically partioning the space allotted to data in clouds. In clouds, space is partitioned evenly across nodes even though certain keys are more popular - their data is requested more often or they have more data than other keys. Christof will work on how to dynamically improve load balance. His other research interests are transport protocols, network simulation, and privacy-preserving communication.

Erik Rodner received his diploma and doctorate at the University of Jena. His thesis work was on the use of machine learning for computer vision and specifically for category recognition. In category recognition, a computer system is asked to label a picture with its category. Erik was interested in reducing the amount of training data a system needs to do that. His system automatically took what it knew from other categories and applied it to the task of recognizing a new category. This allows the system to apply its knowledge about similar categories. He used Bayesian and kernel methods to accomplish this. While at ICSI in the group of Trevor Darrell, he will also work on domain adaptation, which allows a system to transfer what it knows from one domain (such as Google Images) to another (such as Web cam photos). More details about his work are available on his Web page.

Daniel Warneke received his computer science diploma from the University of Paderborn and went on to the doctoral program in computer science at the Berlin Institute of Technology. His thesis was on massively parallel data processing on Infrastructure as a Service platforms. His work examined and improved the efficiency of data-intensive applications on virtualized pay-as-you-go cloud systems like Amazon EC2, for example by exploiting the flexible cost model of those clouds for resource management. Daniel was a member of the Stratosphere research project. He wrote the parallel data processing framework Nephele, which now provides the foundation for the Stratosphere system, a rich open source software stack for next-generation big data analytics.
While at ICSI, Daniel plans to work on resource management for distributed, data-intensive applications.