scispace - formally typeset
Search or ask a question
Author

Mehmet Ali Kutlugun

Bio: Mehmet Ali Kutlugun is an academic researcher from Istanbul Sabahattin Zaim University. The author has contributed to research in topics: Trigram & Speech synthesis. The author has an hindex of 2, co-authored 5 publications receiving 12 citations.

Papers
More filters
Proceedings ArticleDOI
26 Sep 2019
TL;DR: Non-real-time face recognition has been performed by training with new augmented dataset of each picture with many features of a small-scale company, using the Convolutional Neural Network model.
Abstract: Nowadays, deep learning methods have been used in many areas such as big data analysis, speech and image processing with the increasing processing power and the development of graphics processors. In particular, face recognition systems have become one of the most important research topics in biometry. Light direction, reflection, emotional and physical changes in facial expression are the main factors in face recognition systems that make recognition difficult. Training of the system with the available data in small data sets is an important factor that negatively affects the performance. The Convolutional Neural Network (CNN) model is a deep learning architecture used for large amounts of training data. In this study, a small number of employee images set of a small-scale company has been increased by applying different filters. In addition, it has been tried to determine which data augmentation options have more effect on face recognition. Thus, non-real-time face recognition has been performed by training with new augmented dataset of each picture with many features.

10 citations

Proceedings ArticleDOI
02 May 2018
TL;DR: In this study, it is aimed to generate meaningful new Turkish sentences using class-based n-gram model from the sentences in the source data set using a trigram model.
Abstract: Text generation systems provide facilities such as making new information deductions from the existing ones, getting information related to them by going out to a knowledgeable way, and generating more detailed results about the calls to the users by generating the codes entered on the internet. In this study, it is aimed to generate meaningful new Turkish sentences using class-based n-gram model from the sentences in the source data set. In order to realize sentence production, a trigram model is proposed and sentences are generated from the word or word groups in the sentence to the number of groups related to it. Thus, new sentences are generated, none of which were identical to the others.

10 citations

Journal ArticleDOI
TL;DR: In this paper , a method called Graded Similarity Rates (GSR) was proposed to remove data that could disrupt class integrity from a pre-trained deep neural network model.
Abstract: In face recognition systems, light direction, reflection, and emotional and physical changes on the face are some of the main factors that make recognition difficult. Deep metric learning algorithms called representative learning are frequently preferred in this field. However, in addition to the model's success in feature extraction, factors such as the distribution of samples in this database and appropriate classifier preferences also affect the overall performance of the face recognition system. This study it is aimed to create integrity in the database of a pre-trained deep neural network model by obtaining augmented data for classes with a limited number of samples. Thanks to this method called Graded Similarity Rates (GSR), augmented data that could disrupt class integrity has been removed from the database. This way, classes with limited examples are kept integrity, and classifier behavior is used more effectively. The model proposed in the experimental study reached 99.38% accuracy values compared to traditional data augmentation models. Experimental results have shown that the database has an acceptable level of success even at smaller vector sizes and is more organized.

1 citations

Proceedings ArticleDOI
02 May 2018
TL;DR: Text to speech applications are mostly used to extract interaction with the user in high-level multimedia tools and instead of synthesizing textual expressions as monotone, single sound form, it is aimed to be separated into species and sounded as different sound forms.
Abstract: Text to speech applications are mostly used to extract interaction with the user in high-level multimedia tools. These applications usually produce artificial (robotic) sounds. In this study, instead of synthesizing textual expressions as monotone, single sound form, it is aimed to be separated into species and sounded as different sound forms. Briefly, the process of speech synthesis from the text is considered as a text classification problem. Machine learning algorithms have been used to perform this sorting process. As a result of the classification, sound files of correctly classified documents are obtained in the formats initially set as default, and different sound formats are obtained for misclassified documents except for their own category.
Journal ArticleDOI
15 May 2020
TL;DR: In this paper, a Trigram model has been proposed to generate Turkish sentences using class-based n-gram model from the sentences in the source data set and this model was developed for use with rule based approach.
Abstract: Text generation studies are the systems which new knowledge inferences are made by analyzing the existing sentences and meaningful information is obtained from an existing knowledge. These systems provide convenience to users to return more meaningful results related to search results, especially on internet searches. To develop a text generator, there is a need for a linguistic theory to define the sources of natural language and a software tool to process these resources in computer environment. In this study, it is aimed to generate meaningful new Turkish sentences using class-based n-gram model from the sentences in the source data set. Trigram model has been proposed to generate sentences and this model has been developed for use with rule based approach. Unlike other methods, the method used in this study produced meaningful and different sentences with the successive addition method within the framework of the rules determined from the groups divided into triple word groups. Thus, new texts were generated by connecting different sentences from the word or word groups in the source text file as much as the number of the groups that associated with.

Cited by
More filters
Proceedings ArticleDOI
26 Sep 2020
TL;DR: The results show that proposed data augmentation schemes have significantly improved the accuracy of PD recognition on a small dataset using both classical machine learning models and BiLSTM.
Abstract: The lack of dopamine in the human brain is the cause of Parkinson disease (PD) which is a degenerative disorder common globally to older citizens. However, late detection of this disease before the first clinical diagnosis has led to increased mortality rate. Research effort towards the early detection of PD has encountered challenges such as: small dataset size, class imbalance, overfitting, high false detection rate, model complexity, etc. This paper aims to improve early detection of PD using machine learning through data augmentation for very small datasets. We propose using Spline interpolation and Piecewise Cubic Hermite Interpolating Polynomial (Pchip) interpolation methods to generate synthetic data instances. We further investigate on reducing dimensionality of features for effective and real-time classification while considering computational complexity of implementation on real-life mobile phones. For classification we use Bidirectional LSTM (BiLSTM) deep learning network and compare the results with traditional machine learning algorithms like Support Vector Machine (SVM), Decision Tree, Logistic regression, KNN and Ensemble bagged tree. For experimental validation we use the Oxford Parkinson disease dataset with 195 data samples, which we have augmented with 571 synthetic data samples. The results for BiLSTM shows that even with a holdout of 90%, the model was still able to effectively recognize PD with an average accuracy for ten rounds experiment using 22 features as 82.86%, 97.1%, and 96.37% for original, augmented (Spline) and augmented (Pchip) datasets, respectively. Our results show that proposed data augmentation schemes have significantly (p < 0.001) improved the accuracy of PD recognition on a small dataset using both classical machine learning models and BiLSTM

21 citations

Journal ArticleDOI
27 Nov 2020
TL;DR: In this study, the interesting features of Turkish in terms of natural languageprocessing are mentioned and summary info about natural language processing techniques, systems and various sources developed for Turkish are given.
Abstract: Natural language processing is a branch of computer science that combines artificial intelligence with linguistics. It aims to analyze a language element such as writing or speaking with software and convert it into information. Considering that each language has its own grammatical rules and vocabulary diversity, the complexity of the studies in this field is somewhat understandable. For instance, Turkish is a very interesting language in many ways. Examples of this are agglutinative word structure, consonant/vowel harmony, a large number of productive derivational morphemes (practically infinite vocabulary), derivation and syntactic relations, a complex emphasis on vocabulary and phonological rules. In this study, the interesting features of Turkish in terms of natural language processing are mentioned. In addition, summary info about natural language processing techniques, systems and various sources developed for Turkish are given.

3 citations

Journal ArticleDOI
TL;DR: A method to identify the root rot disease in ginseng plants by analyzing the RGB plant images using image processing and deep learning is proposed and the performance of the proposed method is promising.
Abstract: Ginseng is an important medicinal plant in Korea. The roots of the ginseng plant have medicinal properties; thus, it is very important to maintain the quality of ginseng roots. Root rot disease is a major disease that affects the quality of ginseng roots. It is important to predict this disease before it causes severe damage to the plants. Hence, there is a need for a non-destructive method to identify root rot disease in ginseng plants. In this paper, a method to identify the root rot disease by analyzing the RGB plant images using image processing and deep learning is proposed. Initially, plant segmentation is performed, and then the noise regions are removed in the plant images. These images are given as input to the proposed linear deep learning model to identify root rot disease in ginseng plants. Transfer learning models are also applied to these images. The performance of the proposed method is promising in identifying root rot disease.

3 citations

Journal ArticleDOI
TL;DR: This paper created a new Turkish dataset (Tr-D2T) that consists of meaning representation and reference sentence pairs without fine-grained word alignments, which was used to train a sequence-to-sequence neural network for Turkish D2T.

1 citations

Proceedings ArticleDOI
01 Oct 2020
TL;DR: This study has evaluated the application of image manipulation with the Thatcher effect, double illusion, and inversion on the performance of face detection for data augmentation needs where the data obtained has a weakness that is the limited amount of data to create a training model.
Abstract: Image data augmentation is a way that makes it possible to increase the diversity of available data without actually collecting new data. In this study, researchers have evaluated the application of image manipulation with the Thatcher effect, double illusion, and inversion on the performance of face detection for data augmentation needs where the data obtained has a weakness that is the limited amount of data to create a training model. The purpose of this research is to increase the diversity of the data so that it can make predictions correctly if given other similar datasets. To perform face detection on images, it is done using YOLOv3 then comparing the accuracy results from the dataset after and before adding data augmentation.

1 citations