Yusuf Kavurucu

Improving the scalability of ILP-based multi-relational concept discovery system through parallelization

Due to the increase in the amount of relational data that is being collected and the limitations of propositional problem definition in relational domains, multi-relational data mining has arisen to be able to extract patterns from relational data. In order to cope with intractably large search space and still to be able to generate high-quality patterns, ILP-based multi-relational data mining and concept discovery systems employ several search strategies and pattern limitations. Another direction to cope with the large search space is using parallelization. By parallel data mining, improvement in time efficiency and scalability can be provided without further limiting the language patterns. In this work, we describe a method for concept discovery with parallelization on an ILP-based concept discovery system. The non-parallel algorithm, namely Concept Rule Induction System (CRIS), is modified in such a way that the parts that involve high amount of query processing, which causes bottleneck, are reorganized in a data parallel way. The resulting algorithm is called, Parallel CRIS (pCRIS). A set of experiments is conducted in order to evaluate the performance of the proposed method.

Predictive cruise control

Predictive Cruise Control (PCC) is one of the most popular functionality on today vehicles. Briefly, it controls vehicle speed at the desired speed value determined by driver. In almost every vehicle sold today, cruise control could be found because it makes drivability manner easier and besides that it decreases fuel consumption with holding vehicle speed stable. Because of high popularity of cruise control, vehicle companies try to improve cruise control usage and also it is a good way to reduce fuel consumption. Therefore, new functionalities of cruise control become to emerge. One of these is predictive feature of cruise control or shortly PCC. PCC is an optimization problem for reducing fuel consumption and travel time and basically it is about finding the vehicle speed profile on a given slope and traffic profile of the road. Therefore, in this project, a PCC optimization problem is tried to solve with given road slope and traffic profile. Fuel consumption and time based cost functions are used and moreover dynamic programming structure is used for finding solution of optimization algorithm. As solution of the algorithm, vehicle speed profile is visualized with developing graphical user interface at the end of the study.

Performance analysis and improvement of machine learning algorithms for automatic modulation recognition over Rayleigh fading channels

Automatic modulation recognition (AMR) is becoming more important because it is usable in advanced general-purpose communication such as, cognitive radio, as well as, specific applications. Therefore, developments should be made for widely used modulation types; machine learning techniques should be employed for this problem. In this study, we have evaluated performances of different machine learning algorithms for AMR. Specifically, we have evaluated performances of artificial neural networks, support vector machines, random forest tree, k-nearest neighbor, Hoeffding tree, logistic regression, Naive Bayes and Gradient Boosted Regression Tree methods to obtain comparative results. The most preferred feature extraction methods in the literature have been used for a set of modulation types for general-purpose communication. We have considered AWGN and Rayleigh channel models evaluating their recognition performance as well as having made recognition performance improvement over Rayleigh for low SNR values using the reception diversity technique. We have compared their recognition performance in the accuracy metric, and plotted them as well. Furthermore, we have served confusion matrices for some particular experiments.

Graph-based concept discovery in multi relational data

Developments in technology, especially in computer science created the need of storing data in variety of areas. This need created the term database where the data is stored in a useful form. In the database, data is logically integrated in file/files according to relations among them. One of the important issues is to extract knowledge from these databases that hold data in a useful and complete form. This process is called as data mining. The main objective of data mining is to extract implicit and useful knowledge from huge and at first glance meaningless mass of data that is stored in database(s). Multi-Relational databases are the ones in which the data is stored in multiple tables (relations). The relationships between those tables are also stored as tables (relations) in the database. The more effective and commonly known approaches for Multi-Relational Data Mining (MRDM) are based on Inductive Logic Programming (ILP). ILP contains concepts from Inductive Learning and Logic Programming. From this point, the main purpose of MRDM is extracting implicit and trivial knowledge from relational database(s) using ILP approaches and techniques. In this approach, data is represented in graph structures and graph mining techniques are used for knowledge discovery. Concept discovery in multi-relational data mining aims to find relational rules that best describe a relation, called target relation, in terms of other relations in the database, called background knowledge. In this study, a graph-based concept discovery method for concept discovery is presented. The proposed method, namely G-CDS (Graph-based Concept Discovery System), utilizes methods both from substructure-based and path-finding based approaches, hence it can be considered as a hybrid method. G-CDS generates disconnected graph structures for each target relation and its related background knowledge, which are initially stored in a relational database, and utilizes them to guide generation of a summary graph. The summary graph is traversed to find concept descriptors. A set of experiments is conducted on datasets that belong to different learning problems. The experimental results show that G-CDS is capable of learning definitions of target relations that belong to different learning problems.

Evaluation of Machine Learning Algorithms for Automatic Modulation Recognition

Automatic modulation recognition (AMR) becomes more important because of usable in advanced general-purpose communication such as cognitive radio as well as specific applications. Therefore, developments should be made for widely used modulation types; machine learning techniques should be tried for this problem. In this study, we evaluate performance of different machine learning algorithms for AMR. Specifically, we propose nonnegative matrix factorization (NMF) technique and additionally we evaluate performance of artificial neural networks (ANN), support vector machines (SVM), random forest tree, k-nearest neighbor (k-NN), Hoeffding tree, logistic regression and Naive Bayes methods to obtain comparative results. These are most preferred feature extraction methods in the literature and they are used for a set of modulation types for general-purpose communication. We compare their recognition performance in accuracy metric. Additionally, we prepare and donate the first data set to University of California-Machine Learning Repository related with AMR.

A Counting-Based Heuristic for ILP-Based Concept Discovery Systems

Concept discovery systems are concerned with learning definitions of a specific relation in terms of other relations provided as background knowledge. Although such systems have a history of more than 20 years and successful applications in various domains, they are still vulnerable to scalability and efficiency issues —mainly due to large search spaces they build. In this study we propose a heuristic to select a target instance that will lead to smaller search space without sacrificing the accuracy. The proposed heuristic is based on counting the occurrences of constants in the target relation. To evaluate the heuristic, it is implemented as an extension to the concept discovery system called C 2 D. The experimental results show that the modified version of C 2 D builds smaller search space and performs better in terms of running time without any decrease in coverage in comparison to the one without extension.

A comparative study on network motif discovery algorithms

Subgraphs that occur in complex networks with significantly higher frequency than those in randomised networks are called network motifs. Such subgraphs often play important roles in the functioning of those networks. Finding network motifs is a computationally challenging problem. The main difficulties arise from the fact that real networks are large and the size of the search space grows exponentially with increasing network and motif size. Numerous methods have been developed to overcome these challenges. This paper provides a comparative study of the key network motif discovery algorithms in the literature and presents their algorithmic details on an example network.

MPI-based Parallelization for ILP-based Multi-relational Concept Discovery

Multi-relational concept discovery is a predictive learning task that aims to discover descriptions of a target concept in the light of past experiences. Parallelization has emerged as a solution to deal with efficiency and scalability issues relating to large search spaces in concept discovery systems. In this work, we describe a parallelization method for the ILP-based concept discovery system called CRIS. CRIS is modified in such a way that steps involving high query processing are reorganized in a data parallel way. To evaluate the performance of the resulting system, called P-CRIS, a set of experiments is conducted.