Projects

Tübitak-Kariyer Project (2018-2020)

Modeling a Large Vocabulary Turkish Sign Language Recognition System using Deep Learning Methods

The aim of this project is to generate a large vocabulary Turkish Sign Language (TSL) corpus and develop a deep learning based pattern recognition system that can identify the class labels of the signs from isolated sign videos. The system will be designed to work with hard classification examples, which include signs that are similar, in real time and with high classification accuracies. This is a challenging problem; the challenge is in tacking and modeling the interactions of components of a sign that come from multiple channels simultaneously. The research, in this context, will form the basis of a more extended research that we want to carry on on the recognition of a Large Vocabulary Continuous TSL.

AU-BAP Project (2018-2020)

Computational Modeling of the Relationship Between Latent Vector Encoding and Output Image in Generative Adversarial Networks

The aim of this project is to generate a large vocabulary Turkish Sign Language (TSL) corpus and develop a deep learning based pattern recognition system that can identify the class labels of the signs from isolated sign videos. The system will be designed to work with hard classification examples, which include signs that are similar, in real time and with high classification accuracies. This is a challenging problem; the challenge is in tacking and modeling the interactions of components of a sign that come from multiple channels simultaneously. The research, in this context, will form the basis of a more extended research that we want to carry on on the recognition of a Large Vocabulary Continuous TSL.

AU-BAP Project (2015-2016)

Optimization of Part Embedding in Sketches Using Compute Unified Device Architecture (CUDA)

Scope of the project includes parallelization of the part embedding problem for shape grammars on CUDA architecture. There are two primary researches completed successfully during the project: The first one is the parallelization of the cost function (The Edge Strength Function) that is used in the previously developed genetic algorithm based approach. The second one includes the research efforts for the development of a novel approach that is based on Particle Swarm Optimization (PSO) algorithm. The researches performed during the project showed that PSO based optimization gives more effective results than the genetic algorithm based optimization. For this purpose we designed and developed CPU and GPU based solutions to the embedding problem that are based on the PSO based algorithm.

Funded Research Projects:

  • TÜBİTAK ARDEB 3501 (June 2018, June 2020 ): Derin Öğrenme Yöntemleri Kullanılarak Geniş Dağarcıklı Türkçe İşaret Dili Tanıma Sisteminin Modellenmesi (PI)

  • A.Ü. BAP - 18L0443010 (May 2018, May 2020 ): Üretken Çekişmeli Ağlarda Gizli Unsur Kodlayıcı ile Çıktı İmgesi Arasındaki İlişkinin Hesaplamalı Modellenmesi - (PI)

  • A.Ü. BAP - 15H0443009 (2015-2016): Eskizlerde Parça Gömme Probleminin CUDA Mimarisinde Optimizasyon Yoluyla Çözümü - (PI)

  • TÜBİTAK 1000 (2014-2017): Ankara Üniversitesi’nde Nitelikli Araştırma Projesi Üretim Altyapısının Geliştirilmesi - (Researcher)

  • TÜBİTAK 1507 (2011-2013): SOYA: Simulasyon Ortak Yaratım Aracı-(Researcher)

  • T.C. Sanayi ve Ticaret Bakanlığı (2010-2011): Ortak Test - (PI)

  • TÜBİTAK ARDEB 1001 (2008-2010): İki Boyutlu Desenlerde Görsel Düşünme Ve Tasarlama Süreçlerinin Şekil Cebiri Kullanarak Hesaplamalı Modellemesi-(Ph.D. Scholar)

Recent Consultancies to Industry Research Projects:

  • TÜBİTAK TEYDEB 1501 (September 2021, March 2022 ): Yapay Zeka Tabanlı Kasiyersiz Akıllı Market Sistemi Geliştirilmesi (AC)

  • TÜBİTAK TEYDEB 1507 (June 2021-December 2021): Hipofizer Mikroadenomların Derin Öğrenme Tabanlı Bir Yazılım Geliştirilerek Bulunması (AC)