Last Updated:
26/02/2024 - 12:32

  • BAP PhD Thesis Project_TEZ-D-109-2023-11363: Adapting Physics-Informed Neural Networks (PINN) to Multivariate Time Series Models

    Time series mostly have an erratic and complex structure. For this reason, the challenging task in time series analysis is to obtain accurate forecasts of future data from the analysis of the previous records. Deep learning methods are state-of-art techniques that make no assumption about the underlying pattern in the series and are also more robust to noise so they have been one of the most preferred methods for time series forecasting recently. Physics-Informed Neural Network (PINN) is a deep learning method that combines the power of neural networks with the underlying physical laws that govern a particular system. In this study, our objective is to develop PINN method for multivariate time series.

    Project leader: Prof. Dr. Ceylan Yozgatlıgil

    PhD Student: Petek Aydemir

  • BAP General Research Project: : GAP-109-2023-11361: Detecting Stationarity in Time Series Using Artificial Intelligence

    The project aims to develop an artificial intelligence algorithm that can effectively determine whether a given time series is stationary or not, and even identify the reasons behind any non-stationarity, by training on a diverse set of simulated and real-life data comprising both stationary and non-stationary time series of various structures and lengths. Traditional methods for testing stationarity in time series often fall short of accurately capturing the statistical properties, hence the need for a more robust approach. In this study, various artificial neural network architectures will be explored to determine the most effective method. Rather than directly inputting the original time series data into the algorithm, the plan is to utilize techniques such as Wavelet Transform, Variational Autoencoders (VAE), or Convolutional Neural Networks (CNNs) to compress all relevant information about the data into condensed variables. These compressed variables will then feed into the algorithm to train it. By employing this approach, the algorithm will be able to effectively distinguish between stationary

    Projct leader: Prof. Dr. Ceylan Yozgatlıgil

    Researchers: Ozancan Özdemir, Cemre Pınar Yazıcı

  • ARDEB-1001_121E107: Development of a Robotic Mirror Therapy System Based on Motor Learning

    The project will be conducted with a team consisting of researchers from the disciplines of mechanical engineering, neuroscience, statistics, neurology, physical medicine and rehabilitation, and industrial design. The project is also supported by the Center of Excellence in Neuroscience and Neurotechnology, NÖROM ( The proposed project aims to focus on robotic hand rehabilitation targeting hemiplegia patients, specifically addressing the pinch/grasp movement. Goals such as understanding motor relearning mechanisms in the brain during recovery and developing patient-specific models contribute to the academic and scientific depth of the project.

    Project Lead: Assoc. Prof. Dr. Kutluk Bilge Arıkan

    Researcher: Prof. Dr. Ceylan Yozgatlıgil

    Scholarship Holder: Ozancan Özdemir

  • Gap-ODTÜ Project: Application of Machine Learning and Advanced Statistical Methods for Mental Workload Classification in Neuroscience Data (August 2023-Present)

  Project Lead: Fulya Gökalp Yavuz

  Researcher: Serenay Çakar

  You can find the article explaining the starting point of this project here.

  • Project Supported by Other Official Institutions: Modeling the Economic Impacts of Disasters (Index Based) (2020-2022)

   You can access the outputs of this project here and here.

          Student: Ozancan Ozdemir

          Advisor: Prof. Dr. Ceylan Yozgatlıgil