Active Research Projects
Deep Meta-learning based Fault Detection algorithms leveraging multi modal sensor information
One of our recent research topics focuses on deep meta-learning algorithms and sensor fusion techniques within the domain of fault diagnosis. Deep meta-learning aims to train models that can rapidly adapt to new tasks using limited data, making it particularly useful in dynamic environments where faults arise under varying conditions. By utilizing prior knowledge from related tasks, meta-learning algorithms enhance generalization across diverse fault scenarios. Sensor fusion, meanwhile, combines data from multiple sensors to enhance the quality of information for fault diagnosis. This integration mitigates individual sensor limitations, such as noise or incomplete data, by fusing complementary signals. By unifying sensor data into a single representation, meta-learning models are able to capture complex patterns and relationships, thereby improving diagnostic accuracy and facilitating early fault detection. Together, these approaches contribute to the development of robust, adaptive fault diagnosis systems capable of managing the complexities of real-world sensor data.
Artificial Intelligence for 6G communication systems (In Collaboration with Yıldız Technical University funded by TÜBİTAK 1001 program)
While 5G is tasked to transform our lives for the better over the next 10 years, next-generation mobile communications, a.k.a. 6G, will undoubtedly demand even higher energy and spectral efficiencies capable of providing myriads of new services and experience to users everywhere they go.
In this project we contribute into development of Artificial Intelligence based algorithms for next generation communication systems’ problems including Fluid Antenna Multiple Access Systems.
Use of Large Language Models in Time Series Forecasting Problems
Large Language Models (LLMs) have achieved groundbreaking success in the field of natural language processing in recent years and hold the potential to offer innovative solutions for time series forecasting problems. However, effectively aligning the modalities of time series data with natural language to fully leverage the capabilities of these models remains a significant challenge. In this study, these issues will be explored, and potential solutions will be proposed.
Development of Large Language Models for Financial Data Forecasting Problems
Time series forecasting problems have traditionally been a field dominated by machine learning and deep learning techniques. However, in recent years, there has been a significant change in this area with the impact of large language models (LLMs) in different fields. This study aims to investigate how LLMs developed for natural language processing tasks can be adapted for more accurate and efficient prediction of financial time series data. Within the scope of the research, the innovations offered by LLMs in solving complex forecasting problems are analysed in comparison with traditional methods. During this examination, the advantages and limitations of LLMs are evaluated and potential breakthroughs and innovative approaches in the field are explored. As a result, it is aimed to increase the knowledge on the use of LLMs in time series forecasting and to contribute to the development of industry-oriented predictive analytical solutions in this field.
Detection of rare-diseases from DNA sequences by utilizing Artificial Intelligence
This project focuses on developing an advanced framework using Large Language Models (LLMs) to classify DNA sequences associated with rare diseases and predict phenotypic outcomes. Rare diseases are difficult to diagnose due to their complex genetic foundations and limited data. By leveraging transformer-based models like DNABERT and T5, this framework will enhance diagnostic accuracy and deepen the understanding of genotype-phenotype relationships. Data will be sourced from public and proprietary genomic databases, ensuring a diverse and well-annotated dataset, while adhering to GDPR guidelines. Despite challenges like limited data, this approach aims to improve diagnosis, discover new genetic mutations, and advance therapies for rare diseases.
Our collaborators in this project includes:
Bilecik Şeyh Edabali University Molecular Biology and Genetic Department, Turkiye
Eskişehir Technical University Industrial Engineering Department, Turkiye
Cornell University, USA
Georg-August-Universitaet Göttingen, Germany
Development of Federated Learning approaches for Electrical Vehicle fleets state of charge estimation
Link to the main consortium
Past Projects
Development of Intelligent Fault Detection and Predictive Maintenance Systems
In this TÜBİTAK 2232 International Outstanding Researchers project, led by Dr. Eyup Cinar, focuses on creating innovative solutions for IoT systems. As a key outcome of this project, our team has successfully developed an end-to-end scalable IoT platform, alongside designing specialized deep learning algorithms tailored for specific sensor data. This advancement empowers efficient data processing and analytics, enhancing the capabilities of IoT applications.
ECOMAI (Ecological Motor Control with AI)
Worked on the proposal writing with the international partners and provided technical consultancy through project offering (PO) and FPP cycles. The project has successfully passed with high scores.