Elektrik Elektronik Mühendisliği Bölümü Koleksiyonu

Permanent URI for this collectionhttps://hdl.handle.net/20.500.11779/1941

Browse

Now showing 1 - 20 of 34

Citation - WoS: 2
Citation - Scopus: 5
Compositional Neural Network Language Models for Agglutinative Languages
(2016) Saraçlar, Murat; Arısoy, Ebru
Continuous space language models (CSLMs) have been proven to be successful in speech recognition. With proper training of the word embeddings, words that are semantically or syntactically related are expected to be mapped to nearby locations in the continuous space. In agglutinative languages, words are made up of concatenation of stems and suffixes and, as a result, compositional modeling is important. However, when trained on word tokens, CSLMs do not explicitly consider this structure. In this paper, we explore compositional modeling of stems and suffixes in a long short-term memory neural network language model. Our proposed models jointly learn distributed representations for stems and endings (concatenation of suffixes) and predict the probability for stem and ending sequences. Experiments on the Turkish Broadcast news transcription task show that further gains on top of a state-of-theart stem-ending-based n-gram language model can be obtained with the proposed models.
Language Modeling for Turkish Text and Speech Processing
(Springer, 2018) Arısoy, Ebru; Saraçlar, Murat
This chapter presents an overview of language modeling followed by a discussion of the challenges in Turkish language modeling. Sub-lexical units are commonly used to reduce the high out-of-vocabulary (OOV) rates of morphologically rich languages. These units are either obtained by morphological analysis or by unsupervised statistical techniques. For Turkish, the morphological analysis yields word segmentations both at the lexical and surface forms which can be used as sub-lexical language modeling units. Discriminative language models, which outperform generative models for various tasks, allow for easy integration of morphological and syntactic features into language modeling. The chapter provides a review of both generative and discriminative approaches for Turkish language modeling.
Çok Düşük Enerji Tüketen Taşınabilir Kullanıma Uygun Yapay Sinir Ağlarının Donanım Gerçeklemeleri
(2023) Kumbasar, Tufan; Altun, Mustafa; Ayhan, Tuba
Yapay sinir ağları (artificial neural networks, ANN) ile ilgili literatürde yer alan araştırmalar ve bunların endüstriyel uygulamaları son yıllarda hızlı bir şekilde artmaktadır. Buradaki temel motivasyon, geleneksel yöntemler ile yüksek doğruluklu olarak çözülmesi zor problemlerin ANN?ler ile çözülebilmesidir. Diğer taraftan, ANN?lerin kullanımı geleneksel yöntemlere göre, başta enerji olmak üzere, çok daha fazla donanımsal kaynak gerektirmektedir. Örnek vermek gerekirse, 16×16 boyutunda 256 adet piksel içeren oldukça küçük bir görüntünün her bir pikselinin ve ANN ağırlıklarının 8-bitlik girişler ile temsil edildiğini varsayalım. Bu durumda, tek bir yapay nöron, 256 adet 8-bitlik çarpma işlemi, bu çarpım sonuçlarının toplanması için minimum 16-bitlik 255 adet toplama işlemi ve bu toplam sonucunun normalize edilmesi için bir aktivasyon fonksiyonu gerektirir. Görece küçük büyüklükteki bir ANN?de bu nöronlardan yüzlerce olduğu düşünülürse, bu kadar ağırlığın bellekte tutulmasının ve yapılacak aritmetik işlemlerin, özellikle enerji tüketimi açısından, oldukça maliyetli olacağı açıktır. Bu durum ANN?lerin taşınabilir cihazlarda kullanılabilmelerini fazlasıyla kısıtlamaktadır ve bu çalışmanın temel motivasyonlarından biridir. Önerilen çalışmada, çok düşük enerji tüketen ANN?ler önerilen yeni sayı hibrit gösterimi kullanılarak tasarlanmıştır, donanım optimizasyonları yapılmıştır ve nesne takibi uygulamalarında kullanılmıştır. Yapılan çalışmalar aşağıdaki üç ana başlıkta değerlendirilebilir. Bu üç ana başlık çalışmanın desteklediği 119E507 Nolu TÜBİTAK projesinde üç iş paketi olarak yer almaktadır. ? ANN enerji tasarrufu için yeni sayı gösterimlerinin sunulması ve devre bloklarının tasarımının yapılması. ? Enerji odaklı ANN donanım tasarımları ve optimizasyonunun yapılması. ? Nesne takibi yapan ANN tasarımlarının özel tümleşik devreler (application specific integration circuits, ASIC) ve alanda programlanabilir kapı dizileri (field programmable gate arrays, FPGA) tasarım platformlarında gerçeklenmesi.
The Tuned Mass Damper as a Subject in Engineering Mechanics Dynamics
(IEEE, 2022) Dorantes-Gonzalez, Dante Jorge
The course of Engineering Mechanics Dynamics is one of the most challenging courses for both mechanical and civil engineering programs, among others. But few universities dare to introduce projects to enhance students' curiosity, interest, and engagement toward engineering by constructing do-it-yourself physical prototypes, making measurements, and calculations to compete for the best performance. The intention of this project is to introduce a complex multiple-degree-of-freedom vibration problem in an easy manner, namely, the topic of a tuned mass damper (TMD) applied to earthquake-like oscillations. This type of projects directly addresses all seven student outcomes recommended by the Accreditation Board of Engineering and Technology (ABET). The project develops critical thinking and inquiry skills by designing and constructing the prototype of a building-like structure and its corresponding TMD; conducting an experiment under certain constraints to test the attenuation after an initial displacement; applying an open-source freeware to plot and measure underdamped oscillations; calculating main vibration parameters; as well as comparing performance results with another teams. Students approach this complex tunning problem by trial-and-error of key TMD parameters, a strategy that sparks fun and gambling to the process and competition for the best performance in attenuation efficiency. Data from direct observation of students' performance, student surveys, reports, presentation videos, office hours, and interviews showed that students enthusiastically responded at all project stages, understood the TMD functioning, and appreciated the value of dynamics in engineering in a more meaningful way than it would be without this type of projects.
Citation - WoS: 1
Citation - Scopus: 2
Improving the Usage of Subword-Based Units for Turkish Speech Recognition
(IEEE, 2020) Çetinkaya, Gözde; Saraçlar, Murat; Arısoy, Ebru
Subword units are often utilized to achieve better performance in speech recognition because of the high number of observed words in agglutinative languages. In this study, the proper use of subword units is explored in recognition by a reconsideration of details such as silence modeling and position-dependent phones. A modified lexicon by finite-state transducers is implemented to represent the subword units correctly. Also, we experiment with different types of word boundary markers and achieve the best performance by adding a marker both to the left and right side of a subword unit. In our experiments on a Turkish broadcast news dataset, the subword models do outperform word-based models and naive subword implementations. Results show that using proper subword units leads to a relative word error rate (WER) reductions, which is 2.4%, compared with the word level automatic speech recognition (ASR) system for Turkish.
Citation - WoS: 1
Citation - Scopus: 1
Domain Adaptation Approaches for Acoustic Modeling
(IEEE, 2020) Arısoy, Ebru; Fakhan, Enver
In the recent years, with the development of neural network based models, ASR systems have achieved a tremendous performance increase. However, this performance increase mostly depends on the amount of training data and the computational power. In a low-resource data scenario, publicly available datasets can be utilized to overcome data scarcity. Furthermore, using a pre-trained model and adapting it to the in-domain data can help with computational constraint. In this paper we have leveraged two different publicly available datasets and investigate various acoustic model adaptation approaches. We show that 4% word error rate can be achieved using a very limited in-domain data.
Çevrimde Imza Doğrulama için Fpga Üzerinde Gerçek Zamanlı Sistem Tasarımı
(2020) Ayhan, Tuba; Orak, Remzi
Bu proje kapsamında, çevrimde imza doğrulama sistemi gerçeklenmiştir. Sistem dokunmatik ekran üzerinden imza (paraf ya da el yazısı bir karakter) alıp, belleğindeki imza öznitelikleri ile karşılaştırarak imzanın iddia edilen kişiye ait olup olmadığını göstermektedir. Orjinal imza resimleri bellekte tutulmadığından sistem imza hırsızlığına karşı bir miktar dayanıklıdır. Sistem dokunmatik ekran, Zynq-7000 geliştirme kartı ve dokunmatik ekran kaleminden oluşur. İmza atıldıktan 0.13 s sonra doğrulama sonucu ekranda verilir. Kullanım rahatlığı açısından atılan imzanın resmi ekranda da gösterilmektedir. Sistemin test ortamında sınıflama başarımı yetenekli taklitçi için %60 dolayında kalsa da sıradan taklitçi için %100?ü bulmaktadır. Proje kapsamında oluşturulup araştırmacılara açılan veri kümesinde tasniflenmiş 500 imza bulunmaktadır. Projenin tüm kaynak kodları github üzerinden açılmıştır. Proje ile ilgili bilgiler, kodlar, veri kümesi ve kısa video da proje sayfası (https://sites.google.com/mef.edu.tr/imza) üzerinde yayındadır.
Differential Microwave Imaging of Cerebral Hemorrhage Via Dort Method
(IEEE, 2023) Dilman, İsmail; Bilgin, Egemen; Doğu, Semih
Bleeding in the brain tissues may cause fatal health conditions and continuous monitoring of the change in this blood accumulation becomes important in the first few hours after the incident. The continuous post-event monitoring aims to detect the variations in the size and the shape of the hemorrhage regions. To this end, the human head is illuminated by non-ionizing electromagnetic radiation, and the scattered field is measured in different time instants. The decomposition of the time-reversal (DORT) method is then used as the microwave imaging algorithm to produce an indicator function. The performance of the proposed technique is assessed via numerical simulations involving a realistic human head phantom. The results suggest that the DORT method is capable of detecting the changes in multiple simultaneous cerebral hemorrhage regions successfully.
Citation - WoS: 1
Citation - Scopus: 5
Uncertainty-Aware Representations for Spoken Question Answering
(Institute of Electrical and Electronics Engineers Inc., 2021) Ünlü, Merve; Arısoy, Ebru
This paper describes a spoken question answering system that utilizes the uncertainty in automatic speech recognition (ASR) to mitigate the effect of ASR errors on question answering. Spoken question answering is typically performed by transcribing spoken con-tent with an ASR system and then applying text-based question answering methods to the ASR transcriptions. Question answering on spoken documents is more challenging than question answering on text documents since ASR transcriptions can be erroneous and this degrades the system performance. In this paper, we propose integrating confusion networks with word confidence scores into an end-to-end neural network-based question answering system that works on ASR transcriptions. Integration is performed by generating uncertainty-aware embedding representations from confusion networks. The proposed approach improves F1 score in a question answering task developed for spoken lectures by providing tighter integration of ASR and question answering.
Citation - Scopus: 1
Cnn-Based Emotion Recognition Using Data Augmentation and Preprocessing Methods
(Institute of Electrical and Electronics Engineers Inc., 2023) Toktaş, Tolga; Kırbız, Serap; Kayaoğlu, Bora
In this paper, a system that recognizes emotion from human faces is designed using Convolutional Neural Networks (CNN). CNN is known to perform well when trained with a large database. The lack of large and balanced publicly available databases that can be used by deep learning methods for emotion recognition is still a challenge. To overcome this problem, the number of data is increased by merging FER+, CK+ and KDEF databases; and preprocessing is applied to the face images in order to reduce the variations in the database. Data augmentation methods are used to reduce the imbalance in the data distribution that still remains despite the increasing number of data in the merged database. The CNN-based method developed using database merging, image preprocessing and data augmentation, achieved emotion recognition with 80% accuracy.
Regression Analysis of Stock Exchanges During the Ramadan Period: Analysis of 16 Countries
(2016) Tan, A. Serdar; Özlem S.
...
Turkish Speech Recognition
(2018) Arısoy, Ebru; Saraçlar, Murat
Automatic speech recognition (ASR) is one of the most important applications of speech and language processing, as it forms the bridge between spoken and written language processing. This chapter presents an overview of the foundations of ASR, followed by a summary of Turkish language resources for ASR and a review of various Turkish ASR systems. Language resources include acoustic and text corpora as well as linguistic tools such as morphological parsers, morphological disambiguators, and dependency parsers, discussed in more detail in other chapters. Turkish ASR systems vary in the type and amount of data used for building the models. The focus of most of the research for Turkish ASR is the language modeling component covered in Chap. 4.
İnsan-robot Dokunsal (haptik) Etkileşimi için Makine Öğrenme Tabanlı Admitans Kontrolü
(2021) Başdoğan, Çağatay; Patoğlu, Volkan; Niaz, Pouya Pourakbarian; Aydın, Yusuf; Necipoğlu, Serkan; Şirintuna, Doğanay; Çaldıran, Ozan
Yakın gelecekte, fabrika, ev, hastane gibi farklı ortamlarda, insanlar ve robotların birlikte çalışarak, fiziksel etkileşim gerektiren görevleri ortaklaşa yerine getirebilmeleri beklenmektedir. Fiziksel insan-robot etkileşimi konusundaki önemli araştırma konularından birisi, ortaklar arasında doğal bir iletişimin kurulmasıdır. İnsan-robot etkileşimi konusunda hali hazırda çeşitli sayıda çalışmalar bulunmasına rağmen, ortaklar arasındaki fiziksel etkileşimi, bilhassa dokunsal (haptik) tabanlı iletişimi inceleyen çalışmalar sınırlı sayıdadır ve bu tip sistemlerdeki etkileşim hala doğal insan-insan etkileşimine kıyaslandığında yapay kalmaktadır. Bu projede, insanla beraber ortak görevler yapabilecek işbirlikçi bir robot için kesir dereceli ve uyarlamalı (adaptif) bir admitans kontrolcü geliştirildi. Bilgimiz dahilinde kesir dereceli bir admitans kontrolcü insan-robot fiziksel etkileşimi için daha önce kullanılmamıştır. Kesir dereceli kontrolcülerin en önemli özelliği, tamsayı olmayan türev ve integralin kullanılabilmesidir ki bu da bize birleşik sistemin (insan-robot) dinamiğinin modellenmesinde ve denetlenmesinde, tam sayılı bir kontrolcüye göre, esneklik sağlamıştır. Ayrıca, kesir dereceli bir admitans kontrolcünün makine öğrenmesi algoritmaları vasıtasıyla uyarlanabilir şekilde kullanıldığına dair bir örnek literatürde mevcut değildir. Makine öğrenmesi algoritmaları, bizim görev sırasında insanın niyetini anlamamızı ve buna göre görev performansını optimize edecek şekilde kontrolcü parametrelerini seçmemizi sağladı. Projede geliştirilen yöntemlerin etkinliğini sınamak için laboratuvar ortamında, insan ve robot arasında fiziksel etkileşim gerektiren kontrollü deneyler 12 adet denekle yapıldı. Bu deneylerde, denekler, robot koluna bağlanmış bir matkap aracılığıyla dik ve düz tahta bir yüzey üzerinde delikler açtılar. Makina öğrenmesi teknikleri kullanılarak kullanıcın hangi alt-görevi (textit{Bekleme, Serbest Hareket, ve Temas}) yerine getirdiği gerçek zamanlı olarak tespit edildi ve buna göre kontrolcünün parametreleri uyarlandı. Bu sayede, robotun insan tarafından yönlendirilip delik açılacak noktaya yaklaştırılırken (textit{Serbest Hareket}) insana düşük direnç (şeffaflık), delme sırasında (textit{Temas}) ise oluşacak titreşimleri azaltarak sistemi daha kararlı ve güvenli hale getirecek şekilde yüksek direnç göstermesi sağlandı. Bu deneylerden elde edilen sonuçlar, insan-robot etkileşimi için, uyarlamalı ve kesir dereceli bir kontrolcünün tam sayılı ve sabit parametreli bir kontrolcüye göre, görev performanı açısından, çok daha etkili olduğunu gösterdi. Son olarak, projede geliştirilen sistemin endüstriyel ortamda geçerliliğini sınamak için, endüstriyel ortağımız olan As-Metal şirketinden 3 adet işçi laboratuvarımıza davet edildi ve eğrili (curved) bir tahta yüzeyde delik açma deneyleri yapıldı. İşçilerden yüzey üzerinde 3 farklı noktada ve 3 farklı açıda delik açmaları istendi. İşçiler bu görevi yerine getirirken hem işbirlikçi robotumuzdan hem de bir artırılmış gerçeklik arayüzünden destek aldılar. Deneylerden sonra, işçilerden geliştirilen sistem hakkında fikirlerini iletebilecekleri bir anket doldurmaları istendi. Bu anket ve işçilerle yapılan kişisel görüşmeler vasıtasıyla robotun güvenirliği, kullanım kolaylığı ve görevi gerçekleştirmesindeki katkısı ölçüldü. Bu anketten elde edilen sonuçlar bize geliştirilen bu insan-robot etkileşim sisteminin endüstriyel uygulamlar için uygun, kolay, ve etkili olduğunu gösterdi.
Citation - Scopus: 1
A Ran/Sdn Controller Based Connectivity Management Platform for Mobile Service Providers
(Institute of Electrical and Electronics Engineers Inc., 2017) Ayhan, Gökhan; Koca, Melih; Zeydan, Engin; Tan, A. Serdar
In this demo, we demonstrate the integration of radio access network (RAN)/Software-Defined Networking (SDN) controller with a connectivity management platform designed for mobile wireless networks. This is an architecture designed throughout the EU Celtic-Plus project SIGMONA1. OpenDaylight based RAN/SDN controller and the application server are capable of collecting infrastructure and client related parameters from OpenFlow enabled switches and Android based phones respectively. The decision on the best access network selection is computed at the application server using a Multiple Attribute Decision Making (MADM) algorithm and instructed back to Android-based mobile client for execution of access network selection. © 2017 IFIP.
Citation - WoS: 4
Citation - Scopus: 4
Multi-Stream Long Short-Term Memory Neural Network Language Model
(2015) Saraçlar, Murat; Arısoy, Ebru
Long Short-Term Memory (LSTM) neural networks are recurrent neural networks that contain memory units that can store contextual information from past inputs for arbitrary amounts of time. A typical LSTM neural network language model is trained by feeding an input sequence. i.e., a stream of words, to the input layer of the network and the output layer predicts the probability of the next word given the past inputs in the sequence. In this paper we introduce a multi-stream LSTM neural network language model where multiple asynchronous input sequences are fed to the network as parallel streams while predicting the output word sequence. For our experiments, we use a sub-word sequence in addition to a word sequence as the input streams, which allows joint training of the LSTM neural network language model using both information sources.
Live Demo: Design and Fpga Implementation of a Component Level Uav Simulator
(IEEE, 2023) Aydın, Yusuf; Ayhan, Tuba; Akyavaş , İrfan
In this work, we introduce a fast, component based simulation environment for UAVs. The simulator framework is proposed as combination of three sub-models: i. battery, ii. BLDC and propeller, iii. dynamic model. The model parameters are extracted for a particular UAV for testing the simulator. The simulator is implemented on an FPGA to increase simulation speed. The simulator calculates battery SOC, position, velocity and acceleration of the UAV with gravity, drag, propeller air inflow velocity. The simulator runs on the FPGA fabric of XilinxXCKU13P with simulation steps of 1 ms.
Citation - Scopus: 5
A Framework for Automatic Generation of Spoken Question-Answering Data
(Association for Computational Linguistics (ACL), 2022) Manav, Y.; Menevşe, M.Ü.; Özgür, A.; Arısoy, Ebru
This paper describes a framework to automatically generate a spoken question answering (QA) dataset. The framework consists of a question generation (QG) module to generate questions automatically from given text documents, a text-to-speech (TTS) module to convert the text documents into spoken form and an automatic speech recognition (ASR) module to transcribe the spoken content. The final dataset contains question-answer pairs for both the reference text and ASR transcriptions as well as the audio files corresponding to each reference text. For QG and ASR systems we used pre-trained multilingual encoder-decoder transformer models and fine-tuned these models using a limited amount of manually generated QA data and TTS-based speech data, respectively. As a proof of concept, we investigated the proposed framework for Turkish and generated the Turkish Question Answering (TurQuAse) dataset using Wikipedia articles. Manual evaluation of the automatically generated question-answer pairs and QA performance evaluation with state-of-the-art models on TurQuAse show that the proposed framework is efficient for automatically generating spoken QA datasets. To the best of our knowledge, TurQuAse is the first publicly available spoken question answering dataset for Turkish. The proposed framework can be easily extended to other languages where a limited amount of QA data is available. © 2022 Association for Computational Linguistics.
Citation - WoS: 1
Performance of Taiwanese Domestic Equity Funds During Quantitative Easing
(2015) Tan, Ömer Faruk
This study is the first to analyze performance of Taiwanese domestic equity funds between January 2009 and October 2014, the period during which quantitative redirected capital flows toward developing economies and the Taiwanese Stock Exchange Weighted Index compounded at approximately 12.9% annually. Adopting methods endorsed by earlier research, we evaluated 15 Taiwanese equity funds' performance relative to market averages using the Sharpe (1966) and Treynor (1965) ratios and Jensen's alpha method (1968). To test market timing proficiency, we applied the Treynor and Mazuy (1966) and Henriksson and Merton (1981) regression analysis methods. Jensen's alpha method (1968) was used to measure fund managers' stock selection skills. Results revealed that funds significantly under-performed Taiwan's average annual market return and demonstrated no exceptional stock-selection skills and market timing proficiency during the era of quantitative easing.
Citation - WoS: 58
Bidirectional Recurrent Neural Network Language Models for Automatic Speech Recognition
(2015) Chen, Stanley; Sethy, Abhinav; Ramabhadran, Bhuvana; Arısoy, Ebru
Recurrent neural network language models have enjoyed great success in speech recognition, partially due to their ability to model longer-distance context than word n-gram models. In recurrent neural networks (RNNs), contextual information from past inputs is modeled with the help of recurrent connections at the hidden layer, while Long Short-Term Memory (LSTM) neural networks are RNNs that contain units that can store values for arbitrary amounts of time. While conventional unidirectional networks predict outputs from only past inputs, one can build bidirectional networks that also condition on future inputs. In this paper, we propose applying bidirectional RNNs and LSTM neural networks to language modeling for speech recognition. We discuss issues that arise when utilizing bidirectional models for speech, and compare unidirectional and bidirectional models on an English Broadcast News transcription task. We find that bidirectional RNNs significantly outperform unidirectional RNNs, but bidirectional LSTMs do not provide any further gain over their unidirectional counterparts.
Highlighting of Lecture Video Closed Captions
(IEEE, 2020) Yıldırım, Göktuğ; Öztufan, Huseyin Efe; Arısoy, Ebru
The main purpose of this study is to automatically highlight important regions of lecture video subtitles. Even though watching videos is an effective way of learning, the main disadvantage of video-based education is limited interaction between the learner and the video. With the developed system, important regions that are automatically determined in lecture subtitles will be highlighted with the aim of increasing the learner's attention to these regions. In this paper first the lecture videos are converted into text by using an automatic speech recognition system. Then continuous space representations for sentences or word sequences in the transcriptions are generated using Bidirectional Encoder Representations from Transformers (BERT). Important regions of the subtitles are selected using a clustering method based on the similarity of these representations. The developed system is applied to the lecture videos and it is found that using word sequence representations in determining the important regions of subtitles gives higher performance than using sentence representations. This result is encouraging in terms of automatic highlighting of speech recognition outputs where sentence boundaries are not defined explicitly.

Browse

Browsing Elektrik Elektronik Mühendisliği Bölümü Koleksiyonu by Scopus Q "N/A"