TR-Dizin
Permanent URI for this collectionhttps://ada.atilim.edu.tr/handle/123456789/21
Browse
Browsing TR-Dizin by Subject "Bilgisayar Bilimleri"
Now showing 1 - 5 of 5
- Results Per Page
- Sort Options
Article Citation Count: 0Fitting a recurrent dynamical neural network to neural spiking data: tackling the sigmoidal gain function issues(Tubitak Scientific & Technological Research Council Turkey, 2019) Doruk, Reşat Özgür; Doruk, Reşat Özgür; Electrical-Electronics EngineeringThis is a continuation of a recent study (Doruk RO, Zhang K. Fitting of dynamic recurrent neural networkmodels to sensory stimulus-response data. J Biol Phys 2018; 44: 449-469), where a continuous time dynamical recurrentneural network is fitted to neural spiking data. In this research, we address the issues arising from the inclusion ofsigmoidal gain function parameters to the estimation algorithm. The neural spiking data will be obtained from the samemodel as that of Doruk and Zhang, but we propose a different model for identification. This will also be a continuoustime recurrent neural network, but with generic sigmoidal gains. The simulation framework and estimation algorithmsare kept similar to that of Doruk and Zhang so that we can have a solid base to compare the results. We evaluatethe estimation performance in two different ways. First, we compare the firing rate responses of the original and theestimated model. We find that responses of both models to the same stimuli are similar. Secondly, we evaluate variationsof the standard deviations of the estimates against a number of samples and stimulus parameters. They show a similarpattern to that of Doruk and Zhang. We thus conclude that our model serves as a reasonable alternative provided thatfiring rate is the response of interest (to any stimulus).Article Citation Count: 0Mobile Robot Navigation Using Reinforcement Learning in Unknown Environments(2019) Khan, Muhammad Umer; Mechatronics EngineeringIn mobile robotics, navigation is considered as one of the most primary tasks, which becomes more challenging during local navigation when the environment is unknown. Therefore, the robot has to explore utilizing the sensory information. Reinforcement learning (RL), a biologically-inspired learning paradigm, has caught the attention of many as it has the capability to learn autonomously in an unknown environment. However, the randomized behavior of exploration, common in RL, increases computation time and cost, hence making it less appealing for real-world scenarios. This paper proposes an informed-biased softmax regression (iBSR) learning process that introduce a heuristic-based cost function to ensure faster convergence. Here, the action-selection is not considered as a random process, rather, is based on the maximum probability function calculated using softmax regression. Through experimental simulation scenarios for navigation, the strength of the proposed approach is tested and, for comparison and analysis purposes, the iBSR learning process is evaluated against two benchmark algorithms.Article Citation Count: 2A polarity calculation approach for lexicon-based Turkish sentiment analysis(Tubitak Scientific & Technological Research Council Turkey, 2019) Turhan, Çiğdem; Yurtalan, Gökhan; Koyuncu, Murat; Koyuncu, Murat; Turhan, Çiğdem; Information Systems Engineering; Software EngineeringSentiment analysis attempts to resolve the senses or emotions that a writer or speaker intends to send across tothe people about an object or event. It generally uses natural language processing and/or artificial intelligence techniquesfor processing electronic documents and mining the opinion specified in the content. In recent years, researchers haveconducted many successful sentiment analysis studies for the English language which consider many words and wordgroups that set emotion polarities arising from the English grammar structure, and then use datasets to test theirperformance. However, there are only a limited number of studies for the Turkish language, and these studies have lowerperformance results compared to those studies for English. The reasons for this can be incorrect translation of datasetsfrom English into Turkish and ignoring the special grammar structures in the latter. In this study, special Turkish wordsand linguistic constructs which affect the polarity of a sentence are determined with the aid of a Turkish linguist, and anappropriate lexicon-based polarity determination and calculation approach is introduced for this language. The proposedmethodology is tested using different datasets collected from Twitter, and the test results show that the proposed systemachieves better accuracy than the previously developed lexical-based sentiment analysis systems for Turkish. The authorsconclude that especially analysis of word groups increases the overall performance of the system significantly.Article Citation Count: 0Sanal Öğrenme Ortamlarındaki Öğrenci Davranışlarının Kümeleme Yöntemi İle Analiz Edilmesi(2019) Eryılmaz, Meltem; Computer EngineeringÖğrenme sürecini yönetmek ve öğrencilerin kendi öğrenmelerinden sorumlu olmalarını sağlamak amacına yönelik uygun araçlar sunan sanal öğrenme ortamları, bu özellikleri ile eğitimci ve öğrencilere geniş imkanlar sağlamaktadır. Tüm bu imkânlara rağmen aynı öğretim kurumu içerisinde sanal öğrenme ortamlarının yeterince etkili kullanılmadığı görülmektedir. Bu deneysel araştırmada sanal öğrenme ortamı olarak Moodle kullanılan bir üniversitede, kullanıcıların ortamı kullanma bilgileri veri madenciliği tekniklerinden kümeleme algoritmaları ile analiz edilerek, ortamda kullanılması tercih edilmeyen ya da daha az tercih edilen bileşenler tespit edilmiştir. Araştırmaya Ankara Atılım Üniversitesi Fen ve Edebiyat Fakültesi’nde Bilgisayara ve Bilgi Sistemlerine Giriş dersini alan 131 öğrenci katılmıştır. Uygulama sürecinde tüm öğrenciler dersi haftada bir kez yüz yüze geri kalanını Moodle sanal öğrenme ortamında olmak üzere harmanlanmış öğrenme yaklaşımı ile almışlardır. Verilerin analizinde kümeleme algoritmalarından k-ortalamalar analizi kullanılmıştır. Kümeleme analizi ile öğrencilerin sanal öğrenme ortamına yönelik memnuniyetleri, bilgisayar kaygıları ve yıl sonu akademik başarılarına göre nasıl bir dağılım gösterdikleri belirlenmiştir. Ayrıca bu dağılıma göre sanal öğrenme ortamındaki aktivite kullanımları arasında farklılık olup olmadığı değerlendirilmiştir.Article Citation Count: 0Selective word encoding for effective text representation(Tubitak Scientific & Technological Research Council Turkey, 2019) Özkan, Akın; Özkan, Akın; Department of Electrical & Electronics EngineeringDetermining the category of a text document from its semantic content is highly motivated in the literatureand it has been extensively studied in various applications. Also, the compact representation of the text is a fundamental step in achieving precise results for the applications and the studies are generously concentrated to improve itsperformance. In particular, the studies which exploit the aggregation of word-level representations are the mainstreamtechniques used in the problem. In this paper, we tackle text representation to achieve high performance in differenttext classification tasks. Throughout the paper, three critical contributions are presented. First, to encode the wordlevel representations for each text, we adapt a trainable orderless aggregation algorithm to obtain a more discriminativeabstract representation by transforming word vectors to the text-level representation. Second, we propose an effectiveterm-weighting scheme to compute the relative importance of words from the context based on their conjunction with theproblem in an end-to-end learning manner. Third, we present a weighted loss function to mitigate the class-imbalanceproblem between the categories. To evaluate the performance, we collect two distinct datasets as Turkish parliamentrecords (i.e. written speeches of four major political parties including 30731/7683 train and test documents) and newspaper articles (i.e. daily articles of the columnists including 16000/3200 train and test documents) whose data is availableon the web. From the results, the proposed method introduces significant performance improvements to the baselinetechniques (i.e. VLAD and Fisher Vector) and achieves 0.823% and 0.878% true prediction accuracies for the partymembership and the estimation of the category of articles respectively. The performance validates that the proposed contributions (i.e. trainable word-encoding model, trainable term-weighting scheme and weighted loss function) significantlyoutperform the baselines.