Search

Date of Conference: 16-19 May 2016 Conference Name: IEEE 24th Signal Processing and Communications Applications Conference, SIU 2016 In this paper, we study multi-class classification of tweets, where we introduce highly efficient dimensionality reduction techniques suitable for online processing of high dimensional feature vectors generated from freely-worded text. As for the real life case study, we work on tweets in the Turkish language, however, our methods are generic and can be used for other languages as clearly explained in the paper. Since we work on a real life application and the tweets are freely worded, we introduce text correction, normalization and root finding algorithms. Although text processing and classification are highly important due to many applications such as emotion recognition, advertisement selection, etc., online classification and regression algorithms over text are limited due to need for high dimensional vectors to represent natural text inputs. We overcome such limitations by showing that randomized projections and piecewise linear models can be efficiently leveraged to significantly reduce the computational cost for feature vector extraction from the tweets. Hence, we can perform multi-class tweet classification and regression in real time. We demonstrate our results over tweets collected from a real life case study where the tweets are freely-worded, e.g., with emoticons, shortened words, special characters, etc., and are unstructured. We implement several well-known machine learning algorithms as well as novel regression methods and demonstrate that we can significantly reduce the computational complexity with insignificant change in the classification and regression performance.

Date of Conference: 2-5 May 2018 In this paper, we study online nonlinear data regression and propose a highly efficient long short term memory (LSTM) network based architecture. Here, we also introduce on-line training algorithms to learn the parameters of the introduced architecture. We first propose an LSTM based architecture for data regression. To diminish the complexity of this architecture, we use an energy efficient operator (ef-operator) instead of the multiplication operation. We then factorize the matrices of the LSTM network to reduce the total number of parameters to be learned. In order to train the parameters of this structure, we introduce online learning methods based on the exponentiated gradient (EG) and stochastic gradient descent (SGD) algorithms. Experimental results demonstrate considerable performance and efficiency improvements provided by the introduced architecture.

Date of Conference: 2-5 May 2018 In this paper, we investigate online parameter learning for Long Short Term Memory (LSTM) architectures in distributed networks. Here, we first introduce an LSTM based structure for regression. Then, we provide the equations of this structure in a state space form for each node in our network. Using this form, we then learn the parameters via our Distributed Particle Filtering based (DPF) training method. Our training method asymptotically converges to the optimal parameter set provided that we satisfy certain trivial requirements. While achieving this performance, our training method only causes a computational load that is similar to the efficient first order gradient based training methods. Through real life experiments, we show substantial performance gains compared to the conventional methods.

Mobile devices begin to spread increasingly recently to offer a lot of services which personal computers offer. This condition has led to increase in the number of security threats in mobile devices and services. In this paper, it has been made a research on mobile malware and malware detection techniques. Within the scope of the study, a a permission based detection system based on the machine learning methods for Android malware was developed. Developed system is analyzed by using Random Forest, Support Vector Machine and Artificial Neural Networks algorithms.

Nowadays, along with the development of information technologies, storage and analysis of biomedical datasets are easy in health sector. In this area, Machine Learning methods provide a great contribution for evaluation and interpretation of data. In this paper, in addition to Support Vector Machines, Decision Tree, K-Nearest Neighbors, Naive Bayes and Dictionary Learning methods, Random Feature Subspaces (RDL) and Random Instance Subspaces (BDL) methods which are the ensembles of Dictionary Learning are used in biomedical data classification. In the test results, SVM and Dictionary Learning methods, RDL and BDL, which are generated using random feature/instance subspaces achieve optimum accuracy results.

25th Signal Processing and Communications Applications Conference (SIU) -- MAY 15-18, 2017 -- Antalya, TURKEY WOS: 000413813100213 Superpixel based methods have recently shown success in scene segmentation and labeling. In scene labeling, a superpixel algorithm is used first to segment the image into visually consistent small regions; then several feature descriptors are computed and classification is performed for each superpixel. In this paper, Kernel Codebook Encoding (KCB) of superpixel features is proposed. In KCB feature vectors are mapped to multiple codewords in a soft manner, instead of the usual hard quantization. The weights assigned to the codewords are determined by a kernel distance function. KCB method is used for encoding of SIFT features in SuperParsing image parsing algorithm. The developed approach is tested on the SIFT Flow dataset consisting of 2,688 images and 33 classes, and achieves 2.7% increase in parsing accuracy over SuperParsing. Turk Telekom, Arcelik A S, Aselsan, ARGENIT, HAVELSAN, NETAS, Adresgezgini, IEEE Turkey Sect, AVCR Informat Technologies, Cisco, i2i Syst, Integrated Syst & Syst Design, ENOVAS, FiGES Engn, MS Spektral, Istanbul Teknik Univ

Semantic place prediction problem is the process of giving semantic names, such as school, home, office, to locations. Different than the localization problem where the coordinates of a place are predicted, the aim is to semantically characterize the location. While the GPS coordinates of a place can be utilized in solving the problem, phone usage patterns of the users in that location can be used as well. In this paper, the aim is to semantically classify locations of smart phone users utilizing the data collected from wireless interfaces and the motion sensors available on the phones with machine learning algorithms. The efficiency of features extracted from raw data is analysed in terms of metrics such as accuracy, using different classification algorithms. The results show that, while random forest and decision tree algorithms achieve 66% accuracy with only temporal features, adding features from device properties and activity features increases the accuracy up to 99%.

Artificial Bee Colony algorithm, inspired by the foraging behavior of real honey bees, is one of the most important swarm intelligence based optimization algorithms. Like other population based evolutionary computation techniques, Artificial Bee Colony algorithm is suitable for parallelization on distributed architectures. In this paper, we presented a new emigrant creation strategy that is being distributed between subcolonies running simultaneously on the independent compute nodes. The running times and objective function values obtained by the parallelized Artificial Bee Colony algorithm with the proposed model on different number of compute nodes are compared with the sequential counterpart of the algorithm and it is seen that convergence performance of the parallelized Artificial Bee Colony algorithm is significantly improved with the proposed emigrant creation strategy.

24th Signal Processing and Communication Application Conference (SIU) -- MAY 16-19, 2016 -- Zonguldak, TURKEY, WOS: 000391250900110, In this paper, 11 different power quality disturbances were automatically detected by using statistical features with wavelet transform and norm entropy techniques. The best of the created features were selected with forward selection algorithm. Performance of classification algorithms, Support Vector Machines (SVM), Multi Layer Perceptron (MLP), k Nearest Neighbor (KNN) and random subspace KNN (Sub-KNN) which is an ensemble method, were examined. Consequently, the best classification accuracy of 99.3% was achieved by using Sub-KNN and it was appeared that compared to other methods, this algorithm was more robust against the noise., IEEE, Bulent Ecevit Univ, Dept Elect & Elect Engn, Bulent Ecevit Univ, Dept Biomed Engn, Bulent Ecevit Univ, Dept Comp Engn

Compressed sensing theory shows that any signal which is defined as sparse in a given domain can be reconstructed using fewer linear projections instead of using all Nyquist-rate samples. In this paper, we investigate basis pursuit, matching pursuit, orthogonal matching pursuit and compressive sampling matching pursuit algorithms, which are basic compressed sensing based algorithms, and present performance curves in terms of mean squared error for various parameters including signal-to-noise ratio, sparsity and number of measurements with regard to mean squared error. In addition, accuracy of estimation performances has been supported with theoretical lower bounds (Cramer-Rao lower bound and deterministic lower mean squared error). Considering estimation performances, compressive sampling matching pursuit yields the best results unless the signal has a non-sparse structure.