Konstantin’s research interests are in designing new scalable algorithms for data mining and machine learning problems using sampling and sketching techniques. These algorithms summarize massive input data to compact data structures such that many of the key properties of the data can be computed from these summaries. The simplest summarization method is subsampling where the summary consists of a small fraction of the data such that examples are sampled independently at random. However, for many problems subsampling yields poor results and more sophisticated sampling and sketching methods are needed. The algorithms that have resulted from Konstantin’s work yield an approximation of the optimal solution and most of them are randomized, i.e., they are allowed to err with certain probability. At the price of a small loss in accuracy, we can analyze massive datasets for which classic in-memory algorithms require computational resources beyond the capabilities of modern hardware. The size of the summaries is independent or only mildly dependent on the size of the original data, therefore the algorithms scale to massive data.
At LSE Konstantin is working with Professor Milan Vojnovic on the design of novel machine learning algorithms for scalable analysis of network-wide events. These include but are not limited to design of novel graph kernels for anomaly detection and graph classification, and the design of new explicit feature maps for deep learning applications. His research is part of a research collaboration between the Department of Statistics at LSE and Huawei Labs, Paris.
Before joining LSE, Konstantin obtained his PhD from IT University of Copenhagen and worked for NEC Labs Europe. On the personal side, he comes from Sofia, Bulgaria and is an avid football fan.