Péter Kovács

LEMON - an Open Source C++ Graph Template Library

This paper introduces LEMON, a generic open source C++ library providing easy-to-use and efficient implementations of graph and network algorithms and related data structures. The basic design concepts, features, and performance of LEMON are compared with similar software packages, namely BGL (Boost Graph Library) and LEDA. LEMON turned out to be a viable alternative to these widely used libraries, and our benchmarks show that it typically outperforms them in efficiency.

Epileptic Seizure Classification of EEG Time-Series Using Rational Discrete Short-Time Fourier Transform

A system for epileptic seizure detection in electroencephalography (EEG) is described in this paper. One of the challenges is to distinguish rhythmic discharges from nonstationary patterns occurring during seizures. The proposed approach is based on an adaptive and localized time-frequency representation of EEG signals by means of rational functions. The corresponding rational discrete short-time Fourier transform (DSTFT) is a novel feature extraction technique for epileptic EEG data. A multilayer perceptron classifier is fed by the coefficients of the rational DSTFT in order to separate seizure epochs from seizure-free epochs. The effectiveness of the proposed method is compared with several state-of-art feature extraction algorithms used in offline epileptic seizure detection. The results of the comparative evaluations show that the proposed method outperforms competing techniques in terms of classification accuracy. In addition, it provides a compact representation of EEG time-series.

Efficient Heuristics for Maximum Common Substructure Search

Maximum common substructure search is a computationally hard optimization problem with diverse applications in the field of cheminformatics, including similarity search, lead optimization, molecule alignment, and clustering. Most of these applications have strict constraints on running time, so heuristic methods are often preferred. However, the development of an algorithm that is both fast enough and accurate enough for most practical purposes is still a challenge. Moreover, in some applications, the quality of a common substructure depends not only on its size but also on various topological features of the one-to-one atom correspondence it defines. Two state-of-the-art heuristic algorithms for finding maximum common substructures have been implemented at ChemAxon Ltd., and effective heuristics have been developed to improve both their efficiency and the relevance of the atom mappings they provide. The implementations have been thoroughly evaluated and compared with existing solutions (KCOMBU and Indigo). The heuristics have been found to greatly improve the performance and applicability of the algorithms. The purpose of this paper is to introduce the applied methods and present the experimental results.

On application of rational Discrete Short Time Fourier Transform in epileptic seizure classification

This work deals with an adaptive and localized time-frequency representation of time-series signals based on rational functions. The proposed rational Discrete Short Time Fourier Transform (DSTFT) is used for extracting discriminative features in EEG data. We take the advantages of bagging ensemble learning and Alternating Decision Tree (ADTree) classifier to detect the seizure segments in presence of seizure-free segments. The effectiveness of different rational systems is compared with the classical Short Time Fourier Transform (STFT). The comparative study demonstrates that Malmquist-Takenaka rational system outperforms STFT while it can provide a tunable time-frequency representation of the EEG signals and less Mean Square Error (MSE) in the inverse transform.

Column Generation Method for an Agent Scheduling Problem

Abstract This paper discusses a real life problem of daily schedule planning for customer visiting agents. An optimization scheme using a combination of column generation and rounding techniques is proposed for solving this problem. In order to realize an efficient implementation, a polynomial time algorithm is presented for the column generation subproblem. Some technical implementation issues are also discussed and finally, experimental results are shown on real-life problem instances. An implementation of the presented solution is currently in production use by one of the leading contact center service providers of Hungary.

Sleep stage classification using sparse rational decomposition of single channel EEG records

A sparse representation of ID signals is proposed based on time-frequency analysis using Generalized Rational Discrete Short Time Fourier Transform (RDSTFT). First, the signal is decomposed into a set of frequency sub-bands using poles and coefficients of the RDSTFT spectra. Then, the sparsity is obtained by applying the Basis Pursuit (BP) algorithm on these frequency sub-bands. Finally, the total energy of each subband was used to extract features for offline patient-specific sleep stage classification of single channel EEG records. In classification of over 670 hours sleep Electroencephalography of 39 subjects, the overall accuracy of 92.50% on the test set is achieved using random forests (RF) classifier trained on 25% of each sleep record. A comparison with the results of other state-of-art methods demonstrates the effectiveness of the proposed sparse decomposition method in EEG signal analysis.

Making the most of approximate maximum common substructure search

The maximum common substructure (MCS) problem is of great importance in multiple aspects of chemoinformatics. It has diverse applications ranging from lead prediction to automated reaction mapping and visual alignment of similar compounds. Many different algorithms have been developed [1], both exact and approximate. Since the MCS problem is NP-complete, the strict time constraints of most applications can only be realistically satisfied by fast and robust approximation methods. We developed two efficient heuristic algorithms. One is based on the popular approach of reducing the MCS problem to finding the maximum clique in the modular product of the two molecule graphs. The other is based on a new algorithm by Kawabata, called the build-up method [2]. We also incorporated other techniques, for example, the topological fingerprinting primarily used in substructure and similarity searching. We optimized our implementations for use in multiple applications developed at ChemAxon. In some applications, for example, hierarchical MCS-based clustering or similarity search in large databases, the algorithms are required to give close to optimal results in limited time. To meet these conflicting demands, our implementations were enhanced with strong heuristics. Upper bound calculation methods were also applied for screening and early termination purposes. In other applications, for example, reaction mapping or visual alignment, the challenge is that topological features must also be taken into account. Apart from the size of the found common substructure, the determined one-to-one correspondence between the atoms of the molecules is also very important. Effective heuristics were developed to guide the algorithms to prefer those solutions in which the relative positions of the common fragments of the input molecules are as similar as possible. Our implementations have been thoroughly tested and benchmarked. They have also been compared to publicly available solution methods, and integrated into different products at ChemAxon. This has shown that the presented MCS algorithms can adequately cover the conflicting requirements of typical applications. We present our methods and heuristics along with their effects on running time, memory usage, as well as the size and features of the result.