Electrical - Electronic Engineering / Elektrik - Elektronik Mühendisliği

Permanent URI for this collectionhttps://hdl.handle.net/11147/11

Browse

Filters

2009 - 200912010 - 2019242020 - 20212

Settings

Has files: YesInstitution Author: search.filters.institutionAuthor.Karaçalı, Bilge

Search Results

Now showing 1 - 10 of 27
  • Article
    Citation - WoS: 3
    Citation - Scopus: 4
    Quasi-Supervised Strategies for Compound-Protein Interaction Prediction [article]
    (Wiley-VCH Verlag, 2021) Çakı, Onur; Karaçalı, Bilge
    In-silico compound-protein interaction prediction addresses prioritization of drug candidates for experimental biochemical validation because the wet-lab experiments are time-consuming, laborious and costly. Most machine learning methods proposed to that end approach this problem with supervised learning strategies in which known interactions are labeled as positive and the rest are labeled as negative. However, treating all unknown interactions as negative instances may lead to inaccuracies in real practice since some of the unknown interactions are bound to be positive interactions waiting to be identified as such. In this study, we propose to address this problem using the Quasi-Supervised Learning (QSL) algorithm. In this framework, potential interactions are predicted by estimating the overlap between a true positive dataset of compound-protein pairs with known interactions and an unknown dataset of all the remaining compound-protein pairs. The potential interactions are then identified as those in the unknown dataset that overlap with the interacting pairs in the true positive dataset in terms of the associated similarity structure. We also address the class-imbalance problem by modifying the conventional cost function of the QSL algorithm. Experimental results on GPCR and Nuclear Receptor datasets show that the proposed method can identify actual interactions from all possible combinations.
  • Article
    Citation - WoS: 10
    Citation - Scopus: 13
    On the Characterization of Cognitive Tasks Using Activity-Specific Short-Lived Synchronization Between Electroencephalography Channels
    (Elsevier, 2021) Olcay, B. Orkan; Özgören, Murat; Karaçalı, Bilge
    Accurate characterization of brain activity during a cognitive task is challenging due to the dynamically changing and the complex nature of the brain. The majority of the proposed approaches assume stationarity in brain activity and disregard the systematic timing organization among brain regions during cognitive tasks. In this study, we propose a novel cognitive activity recognition method that captures the activity-specific timing parameters from training data that elicits maximal average short-lived pairwise synchronization between electroencephalography signals. We evaluated the characterization power of the activity-specific timing parameter triplets in a motor imagery activity recognition framework. The activity-specific timing parameter triplets consist of latency of the maximally synchronized signal segments from activity onset Delta t, the time lag between maximally synchronized signal segments t, and the duration of the maximally synchronized signal segments w. We used cosine-based similarity, wavelet bi-coherence, phase-locking value, phase coherence value, linearized mutual information, and cross-correntropy to calculate the channel synchronizations at the specific timing parameters. Recognition performances as well as statistical analyses on both BCI Competition-III dataset IVa and PhysioNet Motor Movement/Imagery dataset, indicate that the interchannel short-lived synchronization calculated using activity-specific timing parameter triplets elicit significantly distinct synchronization profiles for different motor imagery tasks and can thus reliably be used for cognitive task recognition purposes. (C) 2021 Elsevier Ltd. All rights reserved.
  • Conference Object
    Citation - WoS: 2
    Türk Makam Müziği Notaları için Otomatik Ezgi Bölütleme
    (Institute of Electrical and Electronics Engineers Inc., 2014) Bozkurt, Barış; Karaçalı, Bilge; Karaosmanoğlu, M. Kemal; Ünal, Erdem
    Automatic melodic segmentation is one of the important steps in computational analysis of melodic content from symbolic data This widely studied research problem has been very rarely considered for Turkish makam music. In this paper we first present test results for state-of-the-art techniques from literature on Turkish makam music data Then, we present a statistical classification-based segmentation system that exploits the link between makant melodies and usul and makam scale hierarchies together with the well-known features in literature. We show through tests on a large dataset that the proposed system has a higher accuracy.
  • Conference Object
    Citation - Scopus: 2
    Model-Free Expectation Maximization for Divisive Hierarchical Clustering of Multicolor Flow Cytometry Data
    (IEEE, 2014) Köktürk, Başak Esin; Karaçalı, Bilge
    This paper proposes a new method for automated clustering of high dimensional datasets. The method is based on a recursive binary division strategy that successively divides an original dataset into distinct clusters. Each binary division is carried out using a model-free expectation maximization scheme that exploits the posterior probability computation capability of the quasi-supervised learning algorithm. The divisions are carried out until a division cost exceeds an adaptively determined limit. Experiment results on synthetic as well as real multi-color flow cytometry datasets showed that the proposed method can accurately capture the prominent clusters without requiring any knowledge on the number of clusters or their distribution models.
  • Conference Object
    Citation - Scopus: 1
    Dijital Sitolojide Kanser Tanıma için Analitik ve Öngörüsel Yarı-güdümlü Öğrenme
    (IEEE, 2012) Karaçalı, Bilge
    In this work, cancer recognition in digital cytology data was carried out using quasi-supervised learning. The data subject to recognition contained ground-truth data only in the form of a labeled set of cancer-free samples and the cancerous samples were provided along with cancer-free samples in an unlabeled mixed dataset. In this framework, a predictive method was derived to label future samples as cancerous or cancer-free based on this data at hand together with an analytical method to label the cancerous samples in the mixed dataset. In the experiments, the methods based on the quasi-supervised learning algorithm achieved higher recognition performance in both cases than the alternative approaches based on supervised support vector machine classifiers. These results indicate that the quasi-supervised learning is the only valid approach in both analytical and predictive recognition when only labeled cancer-free samples are available for statistical learning. © 2012 IEEE.
  • Conference Object
    Citation - Scopus: 1
    Elektroensefalografi Verilerinin Yarı-güdümlü Öğrenme ile Otomatik Olarak İşaretlenmesi
    (IEEE, 2012) Köktürk, Başak Esin; Karaçalı, Bilge
    In this study, the separation of the stimulus effects from the baseline was investigated in electroencephalography data recorded under different visual stimuli using quasi-supervised learning. The data feature vectors were constructed using independent component analysis and wavelet transform, and then, these feature vectors were separated using quasi-supervised learning. Experiment results showed that the EEG data of the stimuli can be separated using quasi-supervised learning. © 2012 IEEE.
  • Conference Object
    Farklı Kontrastlı Medi̇kal Görüntülerde Elasti̇k Hi̇zalama İ̇çi̇n Ni̇rengi̇ Noktalarının Beli̇rlenmesi̇ ve Sınanması
    (IEEE, 2011) Karaçalı, Bilge
    In this study, automatic identification of landmarks was carried out on real double-echo PD/T2 Magnetic Resonance images for use in multi-modality deformable image registration algorithms. To this end, the corner points of the PD-weighted Magnetic Resonance image selected as the reference image were identified at varying fields of view and used as landmark candidates after eliminating the repeated points and those at close proximity. Next, the matching performances of these candidates were determined by invoking a localization algorithm that optimizes information theoretic point similarity measures derived earlier after random initialization at varying distances and directions. In consideration to the best performing point similarity measures and neighborhood radii at landmarks providing successful localization, the mutual information-based point similarity measure evaluated over large neighborhoods was preferred at a conspicuously higher rate than the other alternatives. © 2011 IEEE.
  • Conference Object
    Citation - Scopus: 3
    Gri Seviye Birliktelik Matrisi Öznitelikleri ve Manifold Öğrenme Yardımıyla Histoloji Görüntülerinde Otomatik Doku Sınıflandırılması
    (Institute of Electrical and Electronics Engineers Inc., 2009) Önder, Devrim; Karaçalı, Bilge
    The aim of this work is to perform automated texture classification of histology slides using grayscale images and manifold learning method. Texture feature vectors were obtained using local gray scale co-occurrence matrices and the dimension of the feature vector space was lowered using Isomap dimension reduction. In a lower dimension feature space, k-means clustering operation was performed in order to provide separate texture clusters. In this work, experimental results were obtained using human kidney histology slides. Corresponding feature vectors and determined texture types were given as results.
  • Conference Object
    Citation - Scopus: 2
    Histoloji Görüntülerinde Kanserli Desenlerin Yarı Güdümlü Öğrenme Yöntemiyle Tam Otomatik Sınıflandırılması
    (Institute of Electrical and Electronics Engineers, 2010) Önder, Devrim; Sarıoğlu, Sülen; Karaçalı, Bilge
    The aim of this work is to perform automated texture classification of histology slide images in health and cancerous conditions using quasi-supervised statistical learning method. Tissue images were acquired from histological slides of human colon and were seperated into two groups in terms of normal and disease conditions. Texture feature vectors corresponding to tissue segments of each image were calculated using co-occurrence matrices. Different texture regions were determined by the quasi-supervised statistical learning method using texture features of normal and cancerous groups. ©2010 IEEE.
  • Conference Object
    Citation - Scopus: 1
    Doğrusal Olmayan Gömme Teknikleri Altında Gen Dizilerinin Evrimsel İ̇lişkileri
    (IEEE, 2010) Doğan, Tunca; Karaçalı, Bilge
    We present an error analysis on the application of non-linear embedding on pairwise evolutionary distances inferred over a collection of genetic sequences following multiple sequence alignment. To this end, we have generated gene sequences evolved by random substitutions along three different evolutionary pathways with known evolutionary distances between every sequence pair. We have compared the discrepancy between the inferred evolutionary distances to the true distances before and after non-linear embedding into a low dimensional vector space. The results indicate that non-linear embedding achieves significant reduction in error in the estimated evolutionary distances. Consequently, nonlinear embedding of evolutionary distances can provide more reliable inferences on the evolutionary relationships between genetic sequences. ©2010 IEEE.