Ongoing Projects

In ENSURE, the goal is to understand the dynamics of driving with different types of uncertainty to achieve safe self-driving in complex real world situations. These include uncertainties due to unknown intentions of other agents such as negotiation scenarios at intersections as well as uncertainties due to modelling errors, for example failing to predict future correctly due to an unknown object on the road. The 5-year project prioritizes safety and explainability of self-driving technology to increase its applicability in real world with state of the art deep learning techniques.

Brain-Computer Interfaces (BCIs) enable the leveraging of cerebral activity of users in order to interact with computer systems. Originally designed for assisting muscularly disabled users, a new trend is emerging towards the use of BCI for a larger audience using passive BCI systems, which are able to transparently provide information regarding the users’ mental states. Virtual Reality (VR) technology could largely benefit from inputs provided by passive BCIs. VR enables to immerse users in 3D computer-generated environments, in a way to feel present in the virtual space, allowing through complete control of the environment, to offer several applications ranging from training and education, to social networking and entertainment. Given the growing interest of society and major industrial groups‘ investments, VR is considered as a major revolution in Human-Computer Interaction.
However, to this day, VR has not yet reached its predicted level of democratization and largely remains at the state of an entertaining experiment. This can be explained by the difficulty to characterize users’ mental state during interaction and the inherent lack of adaptation in the presentation of the virtual content. In fact, studies have shown that users experience VR in different ways. While approximately 60% of users experience “cybersickness”, which represents the set of deleterious symptoms that may occur after a prolonged use of virtual reality systems, users can also suffer from breaks in presence and immersion, due to rendering and interaction anomalies which can lead to a poor feeling of embodiment and incarnation towards their virtual avatars. In both cases user’s experience is severely impacted as VR experience strongly relies on the concepts of telepresence and immersion.
The aim of this project is to pave the way to the new generation of VR systems leveraging the electrophysiological activity of the brain through a passive BCI to level-up the immersion in virtual environments. The objective is to provide VR systems with means to evaluate the users’ mental states through the real-time classification of EEG data. This will improve user’s immersion in VR by reducing or preventing cybersickness, and by increasing levels of embodiment through the real time adaptation of the virtual content to the users’ mental states as provided by the BCI.
In order to reach this objective, the proposed methodology is to i) investigate neurophysiological markers associated with early signs of cybersickness, as well as neuromarkers associated with the occurrence of VR anomalies; ii) build on existing signal processing methods for the real-time classification of these markers associating them with corresponding mental states and iii) provide mechanisms for the adaptation of the virtual content to the estimated mental states.

Lorem ipsum dolor sit amet consectetur adipiscing elit. Optio odit, neque molestiae voluptatum illo, ducimus quo cupiditate vitae assumenda ex debitis ullam necessitatibus deleniti sapiente repellendus quae eligendi, corporis eos.

Today, there are approximately 50 million epilepsy patients all over the world, and 30% of these patients are treatment-resistant (WHO, 2022). In this type of epilepsy, called refractory epilepsy, almost none of the patients respond to drug therapy. In addition, it is very difficult to find the source of epileptic seizures in these patients. Although the electroencephalogram (EEG) is the gold standard in the diagnosis of epileptic seizures and is a very successful electrophysiological recording method, its inability to determine the epileptic focus and the visual evaluation of the obtained data only by a properly trained specialist are the most important limitations of this method. To determine the epileptic focus properly is of critical importance, especially for patients who will undergo surgery. In this context, not many, studies in which EEG data are converted into sound (sonification) have been carried out in order to minimize the need for specialists in EEG interpretation and, more importantly, to determine the epileptic focus. In addition to the evaluation of the scalp EEG by conventional methods, the sonification of invasive EEG via intracranial depth electrodes may contribute the precise detection of the epileptic focus. In our project, EEG data will be collected from the scalp and invasive electrodes (ECoG and depth electrodes). Electrode localization will be coregistered depending on the CT images with MRI and CT. Then, multichannel scalp and/or depth electrodes EEGs will be 3D sonificated according to their coordinates and rendered via VR headsets. Within the scope of this project, it is also planned to provide users with the guided sonification feature, which enables the user to be guided to the significant region in terms of epileptic signals, through machine learning as an optional feature.

Hızlı tüketim alışkanlıkları artışı ile yapay zeka teknolojisinin iş entegrasyonun sağlanması birçok inovatif çözümleri beraberinde getirecektir.

Projemizde geçmiş çoklu formülasyon girdileri ve özel kumaşlar ile yapılan yıkamaların spektrofotmetrik ölçümlerin sonuçları ile dijital bir veri kütüphanesi tasarlanarak verileri insana kıyasla etkin ve inovatif kullanacak olan yapay zeka teknolojisine aktarmak hedeflenmektedir. Geliştirilecek olan yapay zeka destekli simülasyon ile çeşitli çamaşır deterjanı formülasyonları oluşturulacak olup, laboratuvarda deney ve performans testi ihtiyacını minimuma düşürecektir. Proje aynı zamanda makine öğrenmesi ile tanımlanan yeni verilere göre öngörü sistemini iyleştirebilir olacaktır.

Proje çıktısı sayesinde beklenen hedefe en yakın formülasyonların hızlı bir şekilde sunulmasıyla kimyasal ve su tüketimini azaltılması ile sürdürülebilirlik kapsamında ve ekonomik fayda sağlayacaktır.

Tofaş Montaj Üretim Müdürlüğü hatlarında, üretim sırasında araç mix değişikliği, araç çeşitliliği ve proses çeşitliliği gibi nedenlerle işçilik kayıpları oluşmaktadır. Araçlar, bulundukları istasyonlarda versiyon ve opsiyon farklılıkları nedeniyle farklı çevrim sürelerine sahiptir.

Projenin amacı, yapay zeka teknolojilerinin kullanımı ile yukarıda bahsedilen tüm sorunları gerçek zamanlı olarak hattan alınan görüntülerle tespit etmek ve ilgililerin süreçleri daha verimli yönetmesini sağlamak üzere analiz oluşturmaktır.

Internet of Energy is an innovative and effective approach that integrates the concept of smart network and Internet technology. Unlike traditional centralized energy systems, distributed Energy Internet system with multiple components and communication requirements needs innovative technologies for reliability and efficiency. Emerging and promising, distributed blockchain, smart contracts, and distributed federated learning technologies offer new opportunities for decentralized Energy Internet systems. Our objective in the SynergyNet project is to develop effective system models, techniques and algorithms by applying innovative distributed blockchain, smart contract and distributed federated learning principles to key research problems and areas within the Energy Internet.   SynergyNet project is funded by TÜBİTAK 2247-A National Research Leaders program research grant. Fully funded PhD student and Postdoctoral researcher positions are available.

BEYONDMOORE addresses the timely research challenge of solving the software side of the Post Moore crisis. The techno-economical model in computing, known as the Moore’s Law, has led to an exceptionally productive era for humanity and numerous scientific discoveries over the past 50+ years. However, due to the fundamental limits in chip manufacturing we are about to mark the end of Moore’s Law and enter a new era of computing where continued performance improvement will likely emerge from extreme heterogeneity. The new systems are expected to bring a diverse set of hardware accelerators and memory technologies. Current solutions to program such systems are host-centric, where the host processor orchestrates the entire execution. This poses major scalability issues and severely limits the types of parallelism that can be exploited. Unless there is a fundamental change in our approach to heterogeneous parallel programming, we risk substantially underutilizing upcoming systems. BEYONDMOORE offers a way out of this programming crisis and proposes an autonomous execution model that is more scalable, flexible, and accelerator-centric by design. In this model, accelerators have autonomy; they compute, collaborate, and communicate with each other without the involvement of the host. The execution model is powered with a rich set of programming abstractions that enable a program to be modeled as a task graph. To efficiently execute this task graph, BEYONDMOORE will develop a software framework that performs static and dynamic optimizations, issues accelerator-initiated data transfers, and reasons about parallel execution strategies that exploit both processor and memory heterogeneity. To aid the optimizations, a comprehensive cost model that characterizes both target applications and emerging architectures will be devised. Complete success of BEYONDMOORE will enable continued progress in computing which in turn will power science and technology in the life after Moore’s Law.

Event camera technology, developed and improved over the past decade, represents a paradigm shift in how we acquire visual data. In contrast to standard cameras, event cameras contain bio-inspired vision sensors that asynchronously respond to relative brightness changes in the scene for each pixel in the camera array and instead produce a sequence of “events” generated at a variable rate at certain times. Hence, they provide very high temporal resolution (in the order of microseconds), high dynamic range, low power consumption and no motion blur. However, because they adopt a fundamentally different design, processing their outputs and unlocking their full potential also require radically new methods. The goal of our project is to contribute to the newly emerging and so-called field of event-based vision.

When compared to event cameras, yet another crucial drawback of traditional cameras is their inability to deal with low-light conditions, which is usually dealt with by employing a longer exposure time in order to allow more light in. This is, however, problematic if the scene to be captured involves dynamic objects or when the camera is in motion, which results in blurry regions. To this end, our project will explore ways to take advantage of event data to improve standard cameras. More specifically, we will investigate enhancing the quality of dark videos as well as accurately estimating optical flow under extremely low-light conditions with the guidance of complementary event data. Toward these goals, we will explore novel deep architectures for constructing intensity images from events and also collect new synthetic and real video datasets to effectively train our models and better test their capabilities.

Our project will provide novel ways to process event data using deep neural networks and will offer hybrid approaches to bring traditional cameras and event cameras together to solve crucial challenges we face when capturing and processing videos in dark. The neural architectures that will be explored in this research project can also be applied to other event-based computer vision tasks. Moreover, as we start to see commercially available high resolution event sensors, we believe that, beyond its scientific impact, our project has also a potential to be commercialized as part of camera systems for future smartphones, mobile robots or autonomous vehicles of any kind.

Inspired by the extraordinary developments achieved in deep learning methods in the last 10 years, there is a paradigm shift in the fields of image/video restoration and compression from traditional analytically engineered methods to nonlinear models learned from data. There are three main factors that affect the performance of learned models: network architecture/capacity, loss function, and size/quality of the data used to train.  We developed new generation image/video restoration and compression solutions by using new visual loss functions, new neuron models and new network architectures to develop explainable models with better generalization properties than their counterparts, and contributed to the international literature in this field.

The original contributions resulting from our theoretical and empirical studies during this project can be summarized as:
• New neuron models for image superresolution and compression: We developed powerful neuron models based on the Pade approximation of nonlinear functions to increase the performance of small-scale models that can be run on mobile devices. The proposed PadeNet architecture based on the~new Pade neurons is applied to the image superresolution task, and its performance is compared with other convolutional network architectures.

VERA aims to develop diagnostic tools for data movement, which is the main source of performance and energy inefficiency in parallel software. Technological advances and big data have increased the importance of data and data has become more critical than computation in terms of both energy consumption and performance in a software. Therefore, there is a need for performance tools that automatically detect and measure data movement in the memory hierarchy and between cores.

Vascular Cognitive Impairment (vCI) is known to be tightly linked to cardiovascular disease (CVD). The main purpose of the CardioStressCI project is to identify and validate causative mechanisms connecting both conditions. We will use an interdisciplinary approach that combines in vitro research with bioinformatics, systems biology modeling and clinical database analysis. We will use network-based disease gene prioritization algorithms to rank the relevance of genes in CI and CVD, and correlate the results to establish interaction networks, which will be modeled using systems biology approaches. Predictions will be validated experimentally with human samples and cell and animal models, to investigate and confirm how individual components of these networks may influence the responses to the different CVD pathological stresses that lead to CI. The main aims of CardioStressCI are: i) to identify proteins linked to CI and CVD; ii) to establish the contribution of nitro-oxidative stress to CVD and CI; iii) to study how CVD induces neuronal dysfunction; iv) to elucidate the pattern of the inflammasome activation in CI and CVD; v) to determine vCI biomarkers. Our studies will consider gender aspects, age, and socio-economic and lifestyle factors as potential modulators of CI pathophysiology. We aim at increasing the knowledge of the molecular mechanisms that contribute to CI when CVD happens. Such knowledge can inform new directions to potentially improve diagnosis, prevention and new therapeutic targets against CI, and even CVD preventing its consequences in brain function.

Sketching is a natural and intuitive means of communication for expressing a concept or an idea. Sketches have various application areas such as problem-solving, design, and art. The increase in the usage of touch and pen-based devices has enabled sketches to be used in human-computer interaction and made sketch recognition an active research area. Our team believes that perception-based models can contribute to the sketch recognition task To be able to integrate perception into our model, we approach the sketch recognition problem with an interdisciplinary perspective.

Our civilization has vitally come to depend on the continuous and affordable scaling of computing performance for consumer electronics, data centers, and high performance computing. Over the past 50 years, the predictable and dependable trend of Moore’s Law has led to incredible innovations and an extremely productive era for humanity. Computers have sequenced the human genome, found new materials and recognized images with greater precision than humans. However, due to fundamental limits in chip manufacturing and the rising energy consumption, this techno-economical model is no longer holding. This incentivizes computer scientists to seek alternatives for continued performance scaling through (i) designing massively parallel architectures and (ii) optimizing software for data movement as it dominates performance and energy inefficiencies. These alternatives expose extra complexities to the end-users who merely wish to focus on their science rather than dealing with architectural intricacies. The goal of this project is to provide diagnostic tools for emerging parallel platforms with a particular focus on data movement as it constitutes the main source of inefficiencies. The main motivation for us is the end-users and hardware designers because the results of this project will substantially shorten the development and optimization cycles of new applications and new hardware. While the outcomes of the project can be used by the end-users (e.g., chemists, data analysts, mechanical engineer, compiler developer) to identify the performance bottlenecks and increase the data locality-awareness of their applications, hardware designers can leverage the diagnostic tools to design more data-centric hardware of the future.

Autonomous vision aims to solve computer vision problems related to autonomous driving. Autonomous vision algorithms achieve impressive results on a single image for various tasks such as object detection and semantic segmentation, however, this success has not been fully extended to video sequences yet. In computer vision, it is commonly acknowledged that video understanding falls years behind single image. This is mainly due to two reasons: processing power required for reasoning across multiple frames and the difficulty of obtaining ground truth for every frame in a sequence, especially for pixel-level tasks such as motion estimation. Based on these observations, there are two likely directions to boost the performance of tasks related to video understanding in autonomous vision: unsupervised learning and object-level reasoning as opposed to pixel-level reasoning. Following these directions, we propose to tackle three relevant problems in video understanding. First, we propose a deep learning method for multi-object tracking on graph structured data. Second, we extend it to joint video object detection and tracking by exploiting temporal cues in order to improve both detection and tracking performance. Third, we propose to learn a background motion model for the static parts of the scene in an unsupervised manner. Our long-term goal is also to be able to learn detection and tracking in an unsupervised manner. Once we achieve these stepping stones, we plan to combine the proposed algorithms into a unified video understanding module and test its performance in comparison to static counterparts as well as the state-of-the-art algorithms in video understanding.

This project will enable self-driving vehicles to deal with rare but catastrophic events, significantly improving their safety and enabling their wide deployment. Self-driving cars are poised to become a trillion-pound market in the next few decades, based on the needs of commuters and logistics chains worldwide. More importantly, they would solve two pressing problems of our society. 1.25 million people die in car accidents each year due to human error, and about 35 million are severely injured — rivaling the worst diseases. Another often ignored fact is that the average car commuter spends 52 minutes per day driving to or from work, amounting to 5.4% of their waking time lost to a menial task. Enabling an Artificial Intelligence (AI) system to understand and drive in complex urban environments now seems largely solved for most common scenarios. Companies such as Waymo (US), Five and Wayve (UK) routinely test on public roads. This achievement was made possible, largely, by advances in deep neural networks. For instance, pedestrian detectors now boast over 98% accuracy, while vehicle position estimation achieves 5 cm errors per 100 meters, according to the widely-acknowledged KITTI benchmark. If self-driving technology is so performant, then why does large-scale deployment keep being delayed? The main reason is safety. Deep networks perform exceedingly well in common cases, but underperform in infrequent driving scenarios. This is a consequence of the fact that they are trained on large datasets by minimizing the empirical risk, which necessarily gives more importance to common cases. We argue that rare events are responsible for the current stalling progress, since they must be individually addressed with manually-defined heuristics or by gathering enough data to make them common. Our goal is to streamline this laborious process. Rather than individually addressing each rare event, we will enable deep networks to do so automatically using one-shot learning. This is a method that trains a deep network specifically to allow efficient adaptation to new scenarios, with only a few examples to learn them. By endowing a network with this ability, and employing data mining techniques to continually discover new rare cases to learn, we will ensure its robustness in a sustainable way. We will rely on two strategic partners whose input will ensure the real-world applicability of our research: Five, one of the most successful self-driving car companies in the UK, and TOGG, Turkey’s national initiative to mass-produce a domestically-developed electric car.

Perfectly aligned with the vision of the EuroHPC Joint Undertaking, the SparCity project aims at creating a supercomputing framework that will provide efficient algorithms and coherent tools specifically designed for maximising the performance and energy efficiency of sparse computations on emerging HPC systems, while also opening up new usage areas for sparse computations in data analytics and deep learning.  The framework enables comprehensive application characterization and modeling, performing synergistic node-level and system-level software optimizations. By creating a digital SuperTwin, the framework is also capable of evaluating existing hardware components and addressing what-if scenarios on emerging architectures and systems in a co-design perspective. To demonstrate the effectiveness, societal impact, and usability of the framework, the SparCity project will enhance the computing scale and energy efficiency of four challenging real-life applications that come from drastically different domains, namely, computational cardiology, social networks, bioinformatics and autonomous driving. By targeting this collection of challenging applications, SparCity will develop world-class, extreme scale and energy-efficient HPC technologies, and contribute to building a sustainable exascale ecosystem and increasing Europe’s competitiveness.

We are experiencing a data-driven revolution at the moment with data being collected at an unprecedented rate. In particular, there is an increasing excitement toward autonomous systems with learning capabilities. Several data-driven applications have already shown significant practical benefit revealing the power of having access to more data, e.g., health care systems, self-driving cars, instant machine translation, and recommendation systems. However, large acceptance of such systems heavily depends on their stability, tractability and reproducibility, where current applications fall inadequate in providing such features. The scale and complexity of modern datasets often render classical data processing techniques infeasible, and therefore, several new algorithms are required to address new technical challenges associated with the nature of the data.

This project focuses on developing efficient and tractable solutions for large-scale learning problems encountered in machine learning and signal processing. Apart from theoretical aspects, the project bears specific goals targeted to applications in principal subspace estimation, low-rank matrix factorization, tensor decomposition and deep learning for largescale systems. Specifically, this novel approach brings together several attractive features:

• The emerging concept of online-learning will be adapted to a distributed setting across a decentralized network topology.
• The exact dynamics of the algorithms will be extracted by a stochastic process analysis method; which current state-of-the-art methods are not able to deliver.
• Studying the extracted dynamics, the learning capabilities and performances of large-scale systems will be improved to match the current needs and challenges of the modern data-driven applications.

The interface between physics and machine learning is a fast-growing field of research. While most studies in this frontier involve using deep learning methods to extract physical knowledge from experimental data or theoretical models, relatively little is known about the nontrivial dynamics of training on artificial neural networks. The analytical framework of nonequilibrium physics developed in the last two decades provides versatile tools that find a novel application in this context. The dynamics of machine learning displays some interesting features not found in physical systems, such as a nontrivial noise structure and a resulting non-thermal steady state, as we observed in our preliminary investigations. The proposed study aims to apply the know-how existing in the nonequilibrium physics literature to this modern problem and explore the implications of various universal laws (originally devised for microscopic systems and expressed in the language of statistical physics) for machine learning. We plan to employ well-known machine learning problems, such as MNIST or CIFAR, as well as some toy models as the testground for analytical predictions. The research team is composed of Dr. Deniz Yuret (Koç Univ) who is an expert in machine learning and the developer of the deep learning package (Knet) for the increasingly popular Julia platform, Dr. Michael Hinczewski (CWRU, USA) who has made important contributions to the literature on nonequilibrium aspects of biological systems, and Dr. Alkan Kabakçıoğlu (Koç Univ, PI), a computational statistical physicist whose recent studies focus on fluctuations and scaling properties of nonequilibrium processes in biomolecules. The proposed research will be conducted in the Department of Physics at Koç University and is expected to last two years.

The researcher Dr. Fatma Guney will carry out a fellowship to create the technology needed to understand the content of videos in a detailed, human-like manner, superseding the current limitations of static image understanding methods, and enabling more robust perception for autonomous driving agents. This fellowship will be carried out at Koc University under the supervision of Prof. Deniz Yuret. Understanding surrounding scene with a detailed and human-level reliability is essential to address complex situations in autonomous driving. State-of-the-art machine vision systems are very good at analysing static images by detecting and segmenting objects on a single image but relating them across time in a video still remains a challenge. Our goal in this proposal is to extend the success of static image understanding to the temporal domain by equipping machines with the ability to interpret videos by exploiting both appearance and motion cues with a human-like ability. There are several challenges which make it difficult for machines such as cluttered backgrounds, the variety and complexity of motion, and partial occlusions due to people interacting with each other.