Location Based Social Networks: Recommendation Generation for the Users [27.11.2019 - 15:45-16:30]


Increasing use of social media and mobile devices lead to the accumulation of more evidence about where people go, what kind of paths they follow, where they are, etc. This evolution led to Location Based Social Networks (LBSN) enabling sharing locations and commenting of locations. Such data that can be obtained from LBSNs enable extraction of patterns about different dimensions of locations and the interaction between people and locations. In this talk, I will focus on generating recommendations for LSBN users, especially context-aware recommendation by using random walk. Additionally, I will talk about recommendations for a group of LBSN users, especially tour recommendations.

Pınar Karagöz (METU)

Dr. Pinar Karagoz is working as full professor in Middle East Technical University (METU) Computer Engineering Department. She received PhD from the same department. During her doctoral studies she worked as a researcher at SUNY Stony Brook University in New York, USA. She had research visits in MIT CSAIL in the USA, Computer Science Department of Ostrava University in Czech Republic, and Computer Engineering Department of Aalto University in Finland. Her research focuses on data mining, machine learning algorithms, information retrieval, social media analysis and mining. She has publications in internationally recognized and indexed international journals including IEEE TKDE, ACM TWEB, IEEE TII and The Computer Journal, and she has about 100 papers in international conferences in the area. In 2016 she received the best paper award in IEEE Transactions on Industrial Informatics. In 2017, her paper was nominated for Wilkes Award of the Computer Journal.

Date: 27.11.2019 - 15:45-16:30

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Fusion and Adaptation for Perception in Intelligent Machines [30.09.2019 - 13:30-14:30]


Intelligent machines are getting more involved in our daily lives. To be able to react autonomously to the environment, they must be able to perceive their surroundings correctly. To deploy multiple sensors on the machine seems to be a proper solution; however, it raises the question of fusion. We develop a hybrid solution to the problem of sensor fusion for autonomous vehicles by using the strengths of two areas: deep learning and Bayesian filters. Bayesian filters are used to predict the dynamic and static regions around the vehicle while deep learning is used to classify the semantics of these regions. We propose semantic grids which is a result of this fusion process. However, even though deep learning methods have a good accuracy in classification, they do not always behave as expected in different environments, such as varying weather conditions. In the second part, I will talk about our unsupervised domain adaptation method. Our method does not only adapt to the weather condition, but it also maintains a high accuracy with the known weather condition. Both qualitative and quantitative evaluations will be shown on commonly used datasets in semantic segmentation for autonomous vehicles.

Özgür Erkent (Team Chroma)

Ozgur Erkent is at the Starting Research Position at Team Chroma. His research interests include robotics, object pose estimation, attentive vision, place recognition, semantic segmentation, object recognition, scene classification, machine learning, human-robot interaction, localization, mechanical and electronical design of robots. He was a Post-doc at Intelligent and Interactive Systems at University of Innsbruck under the supervision of Prof. Dr. Justus Piater from 2013 until 2017 August. He received his PhD in 2013 from Department of Electrical and Electronics Engineering at Boğaziçi University, MSc in 2004 from Cognitive Science Department and BS from Mechanical Engineering in 2001, both from Middle East Technical University.

Date: 30.09.2019 - 13:30-14:30

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Developing a Machine Reasoning Engine for Artificial General Intelligence [23.09.2019 - 10:30-11:30]


Machine learning (ML), in particular deep learning, forms the backbone of the current phase of AI revolution. ML systems can learn to answer questions narrowly posed around the training data sets. The next step in AI revolution towards human-level intelligence will require machine reasoning, or the ability to apply prior knowledge to new situations. We explore to develop a Machine Reasoning (MR) Engine that can learn arbitrary concepts and can discover higher-level cognitive relationships among them. MR Engine will complement the existing ML Systems in three dimensions: 1. It will not require problem-specific model development; 2. It can be trained with Small Data; and 3. It can identify cognitive relationships that are out of reach for ML tools and techniques. MR Engine will be fully parallelizable, where availability of more computing resources will guarantee faster response. This is an ambitious undertaking which can have huge scientific and industrial impact. We will review the building blocks of MR Engine to achieve the above objectives.

Dr. Cengiz Erbas (Amazon)

Dr. Cengiz Erbas is a Software Development Manager at Amazon in Boston, Massachusetts, where he contributes to the development of Amazon's AI Assistant Alexa. Dr. Erbas graduated from the Computer Science and Engineering Department of Hacettepe University. He received his M.Sc. and Ph.D. degrees in Computer Science from Southern Methodist University, Dallas, Texas. He then moved to Silicon Valley and worked for TRW, Wind River and Trimble Navigation where he managed software engineering teams to develop numerous real-time and embedded systems projects. Prior to joining Amazon, Dr. Erbas was with ASELSAN, where he served as the manager of the Image Processing Department at the Microelectronics, Guidance and Electro-Optics division.

Date: 23.09.2019 - 10:30-11:30

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Serious Gaming and Its Applications in Rehabilitation, Cultural Heritage and Training[17.04.2019 - 15:45-16:30]


Serious games, which can briefly be defined as games that have ulterior motives rather than entertainment only, are widely used in various research domains. Education, defense industry and medicine are the main industries which give the misleading “serious” aspect of the game genre. In this talk, definitions and types of serious games will be introduced followed by a brief literature survey. Example games and case studies from recent national and EU-funded projects, which use adaptive difficulty adjustment, procedural content generation and reinforcement learning algorithms, will also be discussed. Finally, opportunities and limitations of using serious games will briefly be mentioned and possible solutions to overcome those limitations will be discussed with feedback from the audience.


Elif Sürer received her Ph.D in Bioengineering in 2011 from the University of Bologna. She received her M.S. and B.S. degrees in Computer Engineering from Boğaziçi University in 2007 and 2005, respectively. From 2013 to 2015 she worked as a post-doctoral researcher in University of Milan in the EU Project REWIRE where she developed video games for the rehabilitation of stroke and Neglect patients. She joined METU Graduate School of Informatics’s Modelling and Simulation Department in 2015 and worked on the training of children having Down Syndrome using video games, which was fully supported by The Scientific and Technological Research Council of Turkey (TÜBİTAK). She is currently working as an Assistant Professor at the METU Graduate School of Informatics’ Multimedia Informatics program. She is funded by the H2020 project eNOTICE as METU local coordinator as of September 2017 and also collaborates as a researcher in several interdisciplinary national and EU-funded projects. She is also a mentor at METU Design Factory and bang. Art Innovation Prix. Her research interests are serious games, virtual/mixed reality, human and canine movement analysis and reinforcement learning.

Date: 17.04.2019 - 15:45-16:30

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Challenges and advances in underwater ad-hoc networks [10.04.2019 - 15:45-16:30]


Underwater networks face many challenges in comparison to terrestrial communication networks due to their unique features. Most of the issues stem from the necessity of using acoustical signals instead of electromagnetic waves to communicate information. One of the main issues is the high propagation delay due to the slow propagation speed of acoustic signals. The high propagation delay makes it difficult to use many of the techniques employed in terrestrial wireless networks in underwater networks. In addition, the mobility of autonomous underwater vehicles (AUV) poses significant challenges for multiple access and routing in ad-hoc AUV networks. Moreover, the difficulty of charging underwater vehicles makes energy consumption a critical issue, especially for battery-powered devices. In this talk, we will provide an overview of such issues encountered in underwater networks and explain some of our proposed solutions to those problems.

Mehmet Köseoğlu (Hacettepe)

Mehmet Köseoğlu received his BS, MS and PhD degrees from Bilkent University, Dept. of Electrical and Electronics Engineering in 2004, 2007 and 2013, respectively. He worked as a software engineer in Aselsan A.Ş between 2004 and 2006. Between 2007 and 2013, he worked as a research assistant in the Department of Electrical and Electronics Engineering at Bilkent University. Since 2015, he has been working as a faculty member in the Department of Computer Engineering at Hacettepe University. He received the Fulbright Postdoctoral Fellowship in 2017 and spent the 2017-2018 academic year as a postdoctoral researcher at the University of California, Los Angeles.

Date: 10.04.2019 - 15:45-16:30

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Data streams: Concepts, a new online ensemble classification algorithm, and a future with no privacy [27.03.2019 - 15:45-16:30]


Data stream classification is a natural part of human skills, and now it can be done more effectively with enormous speed by online machine learning. Ensemble classifiers aim to provide wisdom of crowds with more accurate predictions than that of its individual classifiers. In ensembles the number of components is important for effectiveness and efficiency. How can we combine component decisions in an optimum way? How can we decide the number of components? These interrelated questions are answered for the first time in our research group's recent work. In my talk I will share them with you, and briefly introduce our ongoing research. I will also invite you to consider the implications of data stream processing and look at a future, which we cannot know, of no privacy with manipulated personal choices.

Fazlı Can (Bilkent)

Fazlı Can received the B.S. and M.S. degrees in electrical and electronics and computer engineering and the Ph.D. degree in computer engineering from Middle East Technical University, Ankara, Turkey, in 1976, 1979, and 1985, respectively. He conducted his Ph.D. research under the supervision of Prof. E. Ozkarahan; at Arizona State University, Tempe, AZ, USA, and Intel, Chandler, AZ, USA; as a part of the RAP Database Machine Project. He is currently a faculty member at Bilkent University, Ankara. Before joining Bilkent, he was a tenured full professor at Miami University, Oxford, OH, USA. He co-edited ACM SIGIR Forum from 1995 to 2002 and is a co-founder of the Bilkent Information Retrieval Group, Bilkent University. His interest in dynamic information processing dates back to his 1993 incremental clustering paper in ACM Transactions on Information Systems and some other earlier work with Prof. E. Ozkarahan on dynamic cluster maintenance. His current research interests include information retrieval and data mining.

Date: 27.03.2019 - 15:45-16:30

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Development and Applications of Computational Approaches to Integrate, Analyse and Make Sense out of Large-Scale Biological Data [13.03.2019 - 15:45-16:30]


Artificial learning and data mining techniques are frequently employed to infer meaning from large-scale and noisy biological/biomedical data accumulated in public servers. A key concept in this endeavour is the prediction of the properties of genes and proteins, such as their functions and interactions. Lately, deep learning (DL) based predictive approaches, which outperformed the conventional methods in the fields of computer vision and natural language processing, are started to be applied on biological data to aid the development of novel solutions to the problems in biomedicine. In this talk, I will explain the projects I have been leading for predicting: (i) functions, (ii) small molecule ligands, and (iii) disease related phenotypic implications of proteins in large datasets by developing computational methods. For this, I will introduce our ongoing international research project in the fields of computational drug discovery and repositioning, and biomedical data integration, entitled: "Comprehensive Resource of Biomedical Relations with Deep Learning and Network Representations (CROssBAR)". I will conclude my talk by summarizing the DL-based probabilistic computational method "DEEPscreen" that we have recently developed as a part of the CROssBAR project to identify novel drug candidate molecules to target specific systems/proteins in human.

Tunca Doğan (METU)

Dr. Tunca Doğan has graduated from both the undergraduate and MSc programs in Middle East Technical University, Faculty of Engineering, studying in the fields of food and chemical process engineering. He was introduced to the computational biology and bioinformatics areas during his PhD study back in 2010. He received his joint PhD degree from the interdisciplinary Bioengineering program, hosted by the Electrical & Electronics Engineering Department in Izmir Institute of Technology, and the Graduate School of Health Sciences and the Faculty of Medicine in Dokuz Eylül University in 2013. Dr. Doğan served as a post-doctoral researcher at the University of Cambridge and at the European Bioinformatics Institute in UK between the years 2013 and 2016. Dr. Doğan currently works as a research group leader and as an adjunct faculty member at the Department of Health Informatics’ Bioinformatics track at the Informatics Institute in Middle East Technical University. His research focus can be summarized as developing novel computational methods for biomolecular sequence analysis, protein function prediction and computational drug discovery; using statistical approaches and data mining / artificial learning techniques.

Date: 13.03.2019 - 15:45-16:30

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Color Rendering in the Camera Imaging Pipeline [14.12.2018 - 11:00-12:00]


The camera imaging pipeline consists of a series of routines that process a camera's initial sensor response (i.e., its raw image) to the final RGB image provided to the user. The camera imaging pipeline is embedded in the camera's hardware. This hardware embedding has made it challenging to study individual processing routines. Recently, however, open hardware and software implementations of the camera processing pipeline have been proposed. By taking advantage of these open pipelines, this talk focuses its attention on the "color-rendering" process in the pipeline. Color rendering starts with the estimation of the scene's illumination. The estimated illumination color is used to correct the illumination color cast by performing white-balance, which is an approximation of a full-color constancy correction. Next, the white-balanced colors are mapped to a canonical device-independent perceptual color space. While such color rendering is a core routine on all cameras, there is room for improvement in terms of colorimetric accuracy of this process.

Hakki Can Karaimer

Hakki Can Karaimer is a Computer Science Ph.D. student at the Department of Electrical Engineering and Computer Science of the York University, Toronto. He obtained his M.Sc. degree from Izmir Institute of Technology and B.E. degree from Ege University. His research interests include computer vision, image processing and computational photography.

Date: 14.12.2018 - 11:00-12:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Realizing the IoT vision - In the Intersection of Networks, Devices and Machine Learning [21.11.2018 - 11:00-12:00]


The IoT landscape spreads both deep and wide, intersection a number of different research areas. Which is why this presentation will examine the grand vision of the IoT and the different technologies that exists for realizing this vision. The purpose of this presentation is to give a birds-eye view of the whole area, but at the same time give concrete examples on how the IoT vision will become reality. The aim of this presentation is to create understanding of the grand vision, in order to lift the point of view of the attendees and create an understanding on how all the research details fit into the big picture. Hence, the different parts of realizing an IoT system that fits into this vision will presented, in order to create this understanding. As well as some explanations on how some of our research at Mid Sweden University fits into this vision.

Stefan Forsström

Stefan Forsström received a Master of Science and Engineering from Mid Sweden University in 2009 with his work on location based context aware applications. He continued directly with postgraduate studies and received a Licentiate in Computer and System Sciences in 2011 in which he presented a method for managing information in context aware applications. He defended his Ph.D. thesis in 2014, which focused on distributed IoT services. Today he is an assistant professor in computer engineering and a researcher in the Communication Systems and Networks research group at the Sensible Things that Communicate research center at Mid Sweden University. Today he continues his research on different distributed systems, IoT networks, and smart services.

Date: 21.11.2018 - 11:00-12:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Learning to Recognize and Localize Without Training Examples [23.11.2018 - 11:00-12:00]


In this talk, I will present an overview of our recent work on zero-shot learning problems. In the first part of my talk, I will present our approach for unsupervised zero-shot learning (ZSL) of classes based on their names. We observe that while most existing unsupervised ZSL methods aim to learn a model for directly comparing image features and class names, this can be a difficult task due to dominance of non-visual semantics in underlying vector-space embeddings of class names. To address this issue, we aim to learn a discriminative word representation such that the similarities between class and combination of attribute names fall in line with the visual similarity, without depending on laborious attribute-class relation annotations for unseen classes.
In the second part of my talk, I will summarize our work on fine-grained object recognition and zero-shot learning in remote sensing imagery. For this purpose, we propose a label-embedding based ZSL model that builds on deep convolutional networks as image representation, and, leverage manually annotated attributes, a natural language model, and a scientific taxonomy as auxiliary information. Our experimental results highlight the importance and difficulty of fine-grained recognition in remote sensing, and shows that zero-shot learning can be a promising direction towards building semantically richer models for remote sensing imagery.
In the third part of my talk, I will present our recent on zero-shot object detection. While zero-shot learning has been predominantly studied for image-level object recognition, arguably an equally important problem is the localization of instances of unseen object categories. Towards addressing this problem, we propose a zero-shot object detection model based on adaptation and combination of two mainstream zero-shot image classification approaches within an object detection framework.

Ramazan Gökberk Cinbiş
Middle East Technical University

Ramazan Gökberk Cinbiş graduated from Bilkent University, Turkey, in 2008, and received an M.A. degree from Boston University, USA, in 2010. He was a doctoral student in the LEAR team, at INRIA Grenoble, France, between 2010-2014, and received a PhD degree in computer science from Universite de Grenoble, in 2014. He is currently an assistant professor at Middle East Technical University. His research areas include computer vision and machine learning, with special interest in learning with minimal supervision.

Date: 23.11.2018 - 11:00-12:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Path Planning in Spatial Networks using Time Dependent Obstacle Model [16.11.2018 - 11:00-12:00]


In this talk, path planning algorithms extended to avoid obstacles will be presented together with possible research directions. Optimal path planning in spatial networks is a challenging problem if there are obstacles that change the flow of the traffic on the network. In a time-critical system, time dependent obstacles make a drastic impact on the flow in application areas such as path planning for first responders, emergency response management systems, baggage handling system and routing in wireless networks. Typically, obstacles are modeled as persistent objects in spatial networks and these objects block the flow of traffic and hence they are topologically disregarded in the planning algorithms. However, simply modeling the obstacles as persistent objects is not capable of capturing the realistic scenarios. Temporal behavior of obstacles can be represented as a time dependent model and can be utilized in the path planning algorithms. For that purpose, a spatio-temporal data model is introduced to capture typical types of obstacles that is essential for routing. Based on the model, an obstacle aware path planning algorithm is proposed to avoid obstacles with the waiting option. Under diverse scenarios, the proposed algorithm is shown to achieve promising results that show the potential of the proposed approach in real life.

Engin Demir
Hacettepe University

Engin Demir is an assistant professor in the Computer Engineering Department, Hacettepe University. He has received B.S. in Computer Engineering and Information Science, M.S. and Ph.D. in Computer Engineering from Bilkent University. Following his PhD, he worked as a postdoctoral researcher at the Ohio State University, USA (2009-2011) and the University of Oxford, UK (2011-2012). Before joining Hacettepe University, he worked as an assistant professor in Çankaya University (2016-2018) and in the University of Turkish Aeronautical Association (2012-2016). His research is funded by national and international research agencies such as TUBITAK, NSF, EPSRC, and Microsoft Research. His current research interests include spatial network databases, information retrieval, social network analysis, data science, and algorithmic engineering.

Date: 16.11.2018 - 11:00-12:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

ST-Steiner: Spatio-Temporal Gene Discovery Algorithm for Autism Spectrum Disorder [02.11.2018 - 11:00-12:00]


Whole exome sequencing (WES) studies for Autism Spectrum Disorder (ASD) could identify only around six dozen risk genes to date because the genetic architecture of the disorder is highly complex. To speed the gene discovery process up, a few network-based ASD gene discovery algorithms were proposed. Although these methods use static gene interaction networks, functional clustering of genes is bound to evolve during neurodevelopment and disruptions are likely to have a cascading effect on the future associations. Thus, approaches that disregard the dynamic nature of neurodevelopment are limited. Here, we present a spatio-temporal gene discovery algorithm for ASD, which leverages information from evolving gene coexpression networks of neurodevelopment. The algorithm solves a prize-collecting Steiner forest based problem on coexpression networks, adapted to model neurodevelopment and transfer information from precursor neurodevelopmental windows. The decisions made by thealgorithm can be traced back, adding interpretability to the results. We apply the algorithm on WES data of 3,871 samples and identify risk clusters using BrainSpan coexpression networks of early- and mid-fetal periods. On an independent dataset, we show that incorporation of the temporal dimension increases the predictive power: Predicted clusters are hit more and show higher enrichment in ASD-related functions compared to the state of the art.

Ercüment Çiçek
Bilkent University

Ercüment Çiçek earned his BS (2007) and MS (2009) degrees in Computer Science and Engineering from Sabancı University. He received his Ph.D. degree in Computer Science from Case Western Reserve University in 2013. During his Ph.D., he visited Cold Spring Harbor Laboratory to work on gene discovery algorithms for Autism Spectrum Disorder in 2012. After graduation, he worked as a Lane Fellow in Computational Biology at Carnegie Mellon University till 2015. Since then, he is an Asst. Prof. in the Computer Engineering Department of Bilkent University and is an adjunct faculty member in Computational Biology Department of Carnegie Mellon University. His research is mainly focused on designing machine learning algorithms for analysis of large-scale biological data. He is the recipient of Simons Foundation Autism Research Initiative Explorer Award, TUBITAK Career Award and Turkish Academy of Sciences Outstanding Young Scientist Award.

Date: 02.11.2018 - 11:00-12:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

How can computer science help us understand the neural basis of action perception? [26.10.2018 - 11:00-12:00]


Visual processing of actions is supported by a network of brain regions including occipito-temporal, parietal, and premotor cortex in primates. Although significant progress has been made in understanding the neural basis of action perception in the last two decades, there are still questions whose answers remain unknown. In this talk, I will talk about two fMRI studies with which we address two important questions. In the first study, we utilize computer vision to reveal what aspects of actions are represented in what levels of the action processing network. In the second study, we use clustering techniques to investigate the natural action clusters in the human brain.

Burcu Ayşen Ürgen
Bilkent University

Burcu Ayşen Ürgen is an Assistant Professor at the Department of Psychology, Bilkent University. She is also affiliated with Aysel Sabuncu Brain Research Center and National Magnetic Resonance Research Center (UMRAM). She received her PhD in Cognitive Science from University of California, San Diego (USA) in 2015. Prior to her PhD, she did her BS in Computer Engineering at Bilkent University, and MS in Cognitive Science at Middle East Technical University. Following her PhD, she worked as a postdoctoral researcher at the Department of Neuroscience, University of Parma (Italy), with Professor Guy A. Orban. Dr. Ürgen's primary research area is human visual perception with a focus on biological motion and action perception. In addition to behavioral methods, she uses a wide range of invasive and non-invasive neuroimaging techniques including fMRI, EEG, and intracranial recordings to study the neural basis of visual perception. Her research commonly utilizes state-of-the-art computational techniques including machine learning, computer vision, and effective connectivity. Besides her basic cognitive neuroscience research, Dr. Ürgen also pursues an interdisciplinary research between social robotics and cognitive neuroscience to investigate the human factors that lead to successful interaction with artificial agents such as robots.

Date: 26.10.2018 - 11:00-12:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Network modeling approaches in human diseases [13.04.2018 - 10:00-11:00]


Biology is still moving toward generation of big data. No single data type is capable to fully represent the cellular activity. Bioinformatics is an interdisciplinary field that uses Biology, Computer Science and Engineering to model biological systems to interpret these data better. Beyond the list of molecules from each data type, there is a necessity to consider multiple data jointly and reconstruct the relations between these molecules which implies the necessity of personalized network-based approaches. In my talk, I will describe our integrative network modeling approach called Omics Integrator. Next, I will illustrate an application of Omics Integrator to address a critical question for patient specific therapy in cancer: can network-based approaches reveal patient specific pathways and drug targets? A proof of principle of this strategy has been shown on a group of Glioblastoma multiforme patients (GBM, the most common and aggressive type of malignant human brain tumor) by integrating tumor-specific phosphoproteomic data and known protein-protein interaction network.

Nurcan Tuncbag

Dr. Nurcan Tuncbag received her BS degree in Chemical Engineering from Istanbul Technical University (ITU) in 2005, and her MS and PhD degrees in Computational Science and Engineering from Koc University in 2007 and 2010, respectively. In 2010, she joined the Department of Biological Engineering at Massachusetts Institute of Technology as a postdoctoral associate. In 2014, she joined the Department of Health Informatics at Middle East Technical University. She is the recipient of the FP7 Marie Curie TUBITAK Co-funded Brain Circulation Scheme in 2013, the Young Scientist Award by Science Academy in 2015 and by Parlar Vakfi in 2017. She works in the field of Bioinformatics and her research focus is on revealing how the genome and the networks of interactions among proteins and genes are altered in cells during cancer.

Date: 13.04.2018 - 10:00-11:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Cyclotorsion measurement using scleral blood vessels [06.04.2018 - 10:00-11:00]


Measurements of the cyclotorsional movement of the eye are crucial in refractive surgery procedures. The planned surgery pattern may vary substantially during an operation because of the position and eye movements of the patient. Since these factors affect the outcome of an operation, eye registration methods are applied in order to compensate for errors. While the majority of applications are based on features of the iris, we propose a registration method which uses scleral blood vessels. Unlike previous offline techniques, the proposed method is applicable during surgery. The sensitivity of the proposed registration method is tested on an artificial benchmark dataset involving five eye models and 46,305 instances of eye images. The cyclotorsion angles of the dataset vary between −10° and +10° at 1° intervals. Additionally, a pilot study is carried out using data obtained from a commercially available device. The real data are validated using manual marking by an expert. The results confirm that the proposed method produces a smaller error rate (mean = 0.44 ± 0.41) compared to the existing methods. A further conclusion is that feature extraction algorithms affect the results of the proposed method. The SIFT, SURF64, SURF128 and ASIFT feature extraction algorithms were examined; the ASIFT method was the most successful of these algorithms. Scleral blood vessels are observed to be useful as a feature extraction region due to their textural properties.

Ali Seydi Keçeli

Dr. Ali Seydi Keçeli is a research assistant at the Department of Computer Engineering, Hacettepe University. He received his BSc (2006), MSc (2009) and PhD (2015) degrees from the same department. Before starting his master and PhD studies he worked as software engineer in military projects. His main research areas are image processing, computer vision, data mining, and big data.

Date: 06.04.2018 - 10:00-11:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Automatic Face Analysis and Deep Learning Based Transfer Learning and Domain Adaptation [04.04.2018 - 15:30-16:30]


In this talk, first, I'll provide an overview of our research activities on automatic face analysis conducted at İTÜ SiMiT Lab. Then, I'll talk about some of our recent work on deep learning based transfer learning and domain adaptation.

Hazim Kemal Ekenel
Istanbul Technical University

Dr. Ekenel ( is an Associate Professor at the Department of Computer Engineering in Istanbul Technical University. He received his Ph.D. degree in Computer Science from the University of Karlsruhe (TH) in 2009. He has founded the Facial Image Processing and Analysis group ( at the Department of Computer Science in Karlsruhe Institute of Technology. He was the task leader for face recognition in several large-scale European projects. His face analysis technology has been used by several research labs and companies and has taken part in several demo and press events. He has received the EBF European Biometric Research Award in 2008 for his contributions to the field of face recognition and with the systems they have developed with his team, he received the Best Demo Award at the IEEE International Conference on Automatic Face and Gesture Recognition in 2008. Since January 2011, he has been coordinating the Benchmarking Facial Image Analysis Technologies (BeFIT) initiative and he is the primary organizer of the BeFIT workshops 2011 and 2012.

Date: 04.04.2018 - 15:30-16:30

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Object Detection Through Search with a Foveated Visual System [30.03.2018 - 10:00-11:00]


In this talk, I will present a foveated object detector (FOD) as a biologically-inspired alternative to the sliding window (SW) approach which is the dominant method of search in computer vision object detection. Similar to the human visual system, the FOD has higher resolution at the fovea and lower resolution at the visual periphery. Consequently, more computational resources are allocated at the fovea and relatively fewer at the periphery. The FOD processes the entire scene, uses retino-specific object detection classifiers to guide eye movements, aligns its fovea with regions of interest in the input image and integrates observations across multiple fixations. Our approach combines object detectors from computer vision with a recent model of peripheral pooling regions found at the V1 layer of the human visual system. We assessed various eye movement strategies on the PASCAL VOC 2007 dataset and show that the FOD performs on par with the SW detector while bringing significant computational cost savings.

Emre Akbaş

Dr. Emre Akbas is an assistant professor at the Department of Computer Engineering, Middle East Technical University (METU) since October 2015. Prior to joining METU, he was a postdoctoral research associate at the Department of Psychological and Brain Sciences, University of California Santa Barbara. He received his PhD degree from the Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign in 2011. His BS and MS degrees are from the Department of Computer Engineering, METU.

Date: 30.03.2018 - 10:00-11:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Czecho-slovak-ia-n-iz-at-ion-al-ist-s: Wait what? [23.03.2018 - 10:00-11:00]


Turkish is a morphologically complex language that introduces the sparsity problem in many natural language processing tasks. Morphological segmentation is the task of segmenting a word into its meaningful units that reduces the sparsity in the language. For example, ‘kitapçılarda’ is segmented into following meaningful units: kitap, çı, lar, and da. In this talk, I will mention about two different methods for morphological segmentation that discovers the inner structure of words in a fully unsupervised setting. One method is based on a statistical Bayesian model and the latter one employs a deep learning model.

Burcu Can Buglalilar
Hacettepe University

Burcu Can Buglalilar completed her B.Sc. and M.Sc. in the Department of Computer Engineering, Hacettepe University. She completed her Ph.D. in 2012 in the Computer Science Department, University of York as a member of the Artificial Intelligence group. She was also a visiting researcher in 2014 in University of York. She is mainly interested in natural language processing with the application of unsupervised learning techniques.

Date: 23.03.2018 - 10:00-11:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

A Cost Effective Navigation Approach for Crowd Simulations [09.03.2018 - 10:00-11:00]


Crowd simulation systems have a wide spectrum of application areas including entertainment industry fields such as cinema and video games, and fields such as emergency and evacuation simulations in architectural structures, military simulation systems and urban planning. In the majority of such entertainment applications, crowds are used to create a background within the existing scene. In these simulations named as ambient crowds, there is no need to visualize the individuals forming the crowd with the highest possible quality or to perform costly complex operations to navigate these individuals. Within the scope of my PhD thesis, a new approach has been proposed to simulate these ambient crowds with the lowest possible navigation cost. In this talk, I will introduce three different steering-free navigation methods have been developed for this purpose, give comparison results of them and discuss their contributions briefly.

Öner Barut
Hacettepe University

Dr. Öner Barut is a research assistant at the Department of Computer Engineering, Hacettepe University. He also recieved his bachelor's (2010) and PhD (2017) degree from the same department. His research interests include topics such as computer graphics, crowd simulation and GPGPU computing. His PhD thesis and recent publications specifically focus on navigation in crowd simulations and utilization of specialized GPU hardware.

Date: 09.03.2018 - 10:00-11:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Transition to Third Child - A Machine Learning Perspective [02.03.2018 - 10:00-11:00]


Turkey is approaching the final stage of its fertility transition although its period fertility rate has stagnated around 2.2 per woman in the last decade. The transition from second to third birth is recognized as a critical landmark in completing fertility transition. In particular, we want to investigate the fundamental characteristics of women in such a transition. In this talk, I will show how we employ decision tree modeling, which is a well-known method in machine learning, on the data set of 2013 Turkey Demographic and Health Survey so as to identify mutually exclusive sub-groups of pioneers and laggards in the fertility transition process of Turkey. Overall results of the decision three models suggest that the third-birth propensity in Turkey differs remarkably on the basis of characteristics of women.

Burkay Genç
Hacettepe University

Burkay Genç got his BSc degree from Bilkent University Industrial Engineering Department, his MSc degree from Bilkent University Computer Engineering Department and his PhD from School of Computer Science, University of Waterloo. Before Hacettepe, he worked in Izmir University of Economics as an assistant professor of computer science and as the assistant to the provost; in TED University as an assistant professor of computer science and as the deputy head of the Computer Engineering Department. Now, he works in the Policy and Strategy Studies programme in the Institute of Population Studies. His research focuses on computational geometry, data visualization and data mining. He also teaches in Computer Animation and Game Technologies programme in the Institute of Informatics. He is proudly a learner, a gamer and a Linuxer.

Date: 02.03.2018 - 10:00-11:00

Location: Hacettepe University, Computer Engineering Building, D10 Classroom


ACM Hacettepe'nin Intel ile birlikte düzenlediği AI Workshop 21 Aralık'ta Hacettepe Üniversitesi, Teknokent 1. Arge Binası Konferans Salonunda gerçekleşecek. Etkinlik, makine öğrenmesine bir giriş yapmanızı sağlayacak ve sizi yeni teknolojilerle tanıştıracak.

Katılmak için linkteki formu doldurmanız yeterli :

Broşür 1

Broşür 2

Date: 21.12.2017

Location:Hacettepe Üniversitesi, Teknokent 1. Arge Binası Konferans Salonu

Robotics as a testbed for Computer Science [08.12.2017 - 10:00-11:00]


Robotics is often more associated with the Mechanical and Electrical Engineering disciplines than Computer Science and Engineering (CS), due to historical reasons. However, advances in robotics within the last decade pushes CS to the forefront of this research field.

In this talk, first, I will briefly describe some of the major challenges in robotics that act as realistic testbeds for CS research. I will argue that these challenges, influence and shape the research from a purely CS perspective. I will present results about the perception, learning and use of affordances and their linking with natural language as a case study. Finally, I will conclude by briefly talking about our new projects, and share the problems that we’ve started to tackle.

Erol Şahin
Bilkent University

Erol Şahin received his Ph.D. degree in Cognitive and Neural Systems from Boston University, his M.S. degree in Computer Engineering from METU, and his B.S. degree in Electrical and Electronics Engineering from Bilkent University. He founded the KOVAN Research Lab. at METU Computer Engineering Department in 2002. His research in swarm robotics, cognitive and developmental robotics was awarded 6.5M TL funding from TUBITAK and European Union. By being chosen for the sixth place from 31 projects, he was awarded a 250K Euros worth 53-DOF iCub humanoid robot platform by RobotCub project to support his research in cognitive robotics. He visited the Robotics Institute at Carnegie Mellon University between 2013-2015 as a Marie Curie International Outgoing Fellow. He is the editor of special editions of three journals and five conference/workshop proceedings. He took part as the technical program/area chair in the conferences ANTS'2010, AAMAS'2009 and ICDL/Epirob'2014 and worked as co-chair in the 3. Turkey Robotic Sciences Conference (TORK’2016). He is serving as an Associate Editor for the Adaptive Behavior journal since 2008 and as an Editorial Board member of the Swarm Intelligence journal since 2007.

Date: 08.12.2017 - 10:00-11:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Energy Efficient Architecture for Graph Analytics Accelerators [24.11.2017 - 11:00-12:00]


Specialized hardware accelerators can significantly improve the performance and power efficiency of computing systems. In this presentation, we will focus on hardware accelerators for graph analytics applications and propose a configurable architecture template that is specifically optimized for iterative vertex-centric graph applications with irregular access patterns and asymmetric convergence. The proposed architecture addresses the limitations of the existing multi-core CPU and GPU architectures for these types of applications.

The SystemC-based template we provide can be customized easily for different vertex-centric applications by inserting application-level data structures and functions. After that, a cycle-accurate simulator and RTL can be generated to model the target hardware accelerators. Our experiments showed that the hardware accelerators generated by our template can outperform a 24 core high end server CPU system by up to 3x in terms of performance with up to 65x better power consumption and significantly smaller area.

Özcan Öztürk
Bilkent University

Dr. Öztürk has been on the faculty at Bilkent since 2008 where he currently is an Associate Professor in the Department of Computer Engineering. His research interests are in the areas of manycore accelerators, cloud computing, GPU computing, on-chip multiprocessing, computer architecture, heterogeneous architectures, and compiler optimizations. Prior to joining Bilkent in Spring 2008, he worked as a software optimization engineer in Cellular and Handheld Group at Intel and Marvell. Dr. Ozturk's research has been recognized by Fulbright Senior Scholar Award in 2014, Turk Telekom Research Collaboration Award in 2012, IBM Faculty Award in 2009, European Network of Excellence on High Performance and Embedded Architecture and Compilation (HiPEAC) Paper Award in 2009, and ICPADS Best Paper Award in 2006. He is a member of GSRC, HiPEAC, IEEE, and ACM and is currently serving as editor/reviewer on leading IEEE, ACM, and various other journals. Dr. Öztürk received his Ph.D. degree in computer science and engineering from the Pennsylvania State University, the M.S. degree in computer engineering from University Of Florida, and received the B.Sc. degree in computer engineering from Bogazici University, Turkey.

Date: 24.11.2017 - 11:00-12:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

A Hybrid Assessment Approach for Medical Device Development Companies [23.11.2017 - 10:00-11:00]


Medical device software development organisations are bound by regulatory requirements and constraints to ensure that developed medical devices will not harm patients. Medical devices have to be treated as complete systems and be evaluated in this manner. Medical device manufacturers in the US as well as in the EU must satisfy the associated regulatory demands of the region that the device will be marketed in. A variety of international standards have to be implemented in order to ensure regulatory requirements for a medical device. Instead of manufacturers having to ensure compliance to various regulatory standards individually, the DkIT researchers previously developed a medical device software process assessment framework called MDevSPICE® that integrates the regulatory requirements from all the relevant medical device software standards. MDevSPICE® was developed in a manner that suits plan-driven software development. In order to improve the usability of MDevSPICE in agile settings, a hybrid assessment approach has been developed.

The hybrid assessment approach combines the MDevSPICE® based process assessment method with steps for prioritization of improvement needs through value stream mapping, and enabling process improvement through the use of the KATA technique. This approach integrates agile methods into the medical device software development process whilst adhering to the requirements of the regulatory standards. The purpose of this seminar is to describe the approach and its implementation within four organisations that develop software in line with medical device regulations in Ireland.

Özden Özcan-Top
(Dundalk Institute of Technology, Ireland)

Dr. Özden Özcan-Top is a post-doctoral researcher at Dundalk Institute of Technology (DkIT), Ireland. Her primary research area is improving software quality and efficiency in the safety critical domain, specifically in medical device software development through the integration of agile software development methods. She received her M.Sc. and PhD degrees from Information Systems (IS) Department at Informatics Institute of Middle East Technical University (METU) in 2008 and 2015 respectively. She received her bachelor’s degree from Industrial Engineering Department at Yıldız Technical University in 2005. She worked as a research assistant at IS Department, METU for 4 years. Following that she worked at Fujitsu Technology Solutions Company as a quality specialist for 2,5 years and obtained hands-on practice on the topics that she had been performing research. She was involved in three TÜBİTAK projects as a researcher and published more than 20 artifacts. Her other research interests include software process assessment/improvement, software project management and software measurement and estimation.

Date: 23.11.2017 - 10:00-11:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Global Path Planning Strategies for Crowd Navigation [27.10.2017 - 10:00-11:00]


There are applications of crowd simulation systems in entertainment industry fields such as cinema and video games, and moreover in fields such as emergency and evacuation simulations in architectural structures, military simulation systems and urban planning. In the context of crowd navigation, the local navigation methods consider each agent in the crowd individually and plan their short term movements, while the global navigation methods approach the crowd as a whole. The hybrid navigation methods, on the other hand, aim to combine the successful parts of these two types that have been developed. Although hybrid navigation methods are more costly as they need to do more calculations, provided that the cost can be kept at a reasonable degree, they can produce much more successful results than local and global methods. In this talk I will explain two hybrid methods I have developed as a part of my PhD thesis, give a brief comparison of them and discuss their contributions on local navigation.

Cumhur Yiğit Özcan
Hacettepe University

Dr. Cumhur Yiğit Özcan is a research assistant at the Department of Computer Engineering, Hacettepe University. He also received his bachelor's (2010) and PhD (2017) degree from the same department. His research interests include topics such as computer graphics, AI for video games and crowd simulations. His PhD thesis and recent publications specifically focus on navigation and path planning in crowd simulations.

Date: 27.10.2017 - 10:00-11:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Keyphrase Extraction and Topic Segmentation [13.10.2017 - 10:00-11:00]


When we express some idea or story, it is inevitable to use words that are semantically related to each other. When this phenomenon is exploited from the aspect of words in the language, it is possible to infer the level of semantic relationship between words by observing their distribution and use in discourse. From the aspect of discourse, it is possible to model the structure of the document by observing the changes in the lexical cohesion in order to attack high-level natural language processing tasks. In this talk lexical cohesion is investigated from both of these aspects by first building methods for measuring semantic relatedness of word pairs and then using these methods in the tasks of topic segmentation and keyphrase extraction.

Gönenç Ercan
Hacettepe University

Gonenc Ercan is a faculty in Informatics Institute of Hacettepe University. Prior to joining Hacettepe University, he has served in Computer Engineering Department of Cankaya University as an Assistant Professor. He received his Ph.D. and M.Sc. degrees from Bilkent University and B.Sc. from Eastern Mediterranean Universtiy.

His research interests are Text Mining, Natural language processing and Information Retrieval focusing on the use of Lexical Semantics in high-level tasks. He has published articles in Information Science journals and conferences.

Date: 13.10.2017 - 10:00-11:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Energy Efficient Smart Programming Languages and Data Centers [29.09.2017 - 10:00-11:00]


As the size of the data centers and supercomputers keeps growing with the computational demand, so too does their power consumption. Current systems consume tens of megawatts of power daily, leading to millions of dollars in energy bills, significant power strains on local, state, or regional energy grid systems, and the environmental impact on natural resources that provide the power for the system. In the near future, as the computational system scales grow, the power costs might increase to hundreds of megawatts daily. This talk will explain the infrastructure of data centers and novel energy efficiency techniques using smart and dynamic programming languages that have been analyzed on top supercomputers in the world using thousands of processors.

Bilge Acun
Data Centric Systems team at IBM T.J. Watson Research Center in New York

Bilge Acun is a Research Scientist at the Data Centric Systems team at IBM T.J. Watson Research Center in New York. She received her Ph.D. from the Department of Computer Science at University of Illinois at Urbana-Champaign (UIUC) and received a B.S. from Bilkent University with magna cum laude.

Bilge's research interests are parallel and distributed computing, green computing, and high performance computing. Her research has been recognized as a cover-featured article in the prestigious IEEE Computer, as well as at top conferences in the area of High Performance Computing. She is the lead inventor of three patent-pending technologies.

Date: 29.09.2017 - 10:00-11:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Legal Information Retrieval: Unification of Legal Precedents [April 7, 2017 - 10:00-10:50]


Legal professionals are ethically obligated to be reasonably informed on legal documents related with the jurisdictions. Digitization of legal documents such as legislations, case laws, and scholarly work made it possible to search for relevant documents. Indeed, legal documents are an important source of data that has different characteristics compared with the typical textual documents. The jargon, polysemes and frequent changes in the domain make it harder to directly apply traditional information retrieval techniques. Previous applied techniques (i.e., keyword base search (full text search), classification and natural language processing with the utilization of legal ontologies) have been applied in the legal domain to achieve high recall (increasing the number of relevant documents) and high precision (reducing the number of irrelevant documents) but due to the nature of the domain failing to retrieve an important or precedential case is yet a serious problem. As a case study, unification of legal precedents is introduced. The similarity and relatedness of the documents are modeled as a network. It has been shown that maximal cliques in the network presents a promising grouping among the legal precedents.

Engin Demir
Çankaya University

Engin Demir is currently an assistant professor in the Computer Engineering Department of Çankaya University. He has received his B.S. in Computer Engineering and Information Science, M.S. and Ph.D. degrees in Computer Engineering from Bilkent University. Before joining Çankaya University, he was an assistant professor in the University of Turkish Aeronautical Association (THKU), served as the head of department between 2012 and 2016. He had worked as a postdoctoral researcher at the Ohio State University, USA (2009-2011) and the University of Oxford, UK (2011-2012). He has worked in several national and international research projects, funded by TUBITAK, NSF, EPSRC, Microsoft Research. His research interests include information retrieval, social network analysis, data mining, spatial network databases, algorithmic engineering.

Date: 07.04.2017 - 10:00-10:50

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Machine Learning in Adverse Settings [29.03.2017 - 10:00-10:50]


Machine learning and signal processing algorithms are finding use in diverse applications such as recommendation generation, surveillance and neural signal analysis. The data in these settings are typically voluminous and noisy. We propose computationally scalable novel algorithms for processing in adverse settings. Of special interest to us are the challenges of non-stationarity, abnormality and low SNR. We first introduce a highly dynamical online classification algorithm with strong mathematical guarantees regardless of source statistics. Our algorithm combines low-complexity methods to achieve superior modeling performance and adaptation to non-stationarity. We demonstrate the approach across several computer vision applications to object recognition, target tracking and anomaly detection for surveillance. Next, we address a problem in the domain of neural signal analysis that illustrates the challenge of interpreting signals with low SNR. Specifically, we attempt to identify the neural bases of naturalistic human face perception. Our proposed solution involves searching for statistical regularities in the data to find robust spatial, temporal as well as spectral features. These are combined through boosting to classify the stimuli as face or non-face in both dynamic and static conditions. Our results yield clues regarding the neural substrates of face processing in the human brain and also have implications for the applied areas of brain-machine interfaces and neurological diagnostics.

Huseyin Ozkan
Massachusetts Institute of Technology

Dr. Huseyin Ozkan is a postdoctoral research associate in Vision and Computational Neuroscience at Massachusetts Institute of Technology. He received his B.Sc. degrees in Electrical Engineering and Mathematics from Bogazici University; and his M.Sc. and Ph.D. degrees in Electrical Engineering from Boston University, and from Bilkent University, respectively. His research interests are in machine learning, signal processing, computer vision and computational neuroscience. He also worked full-time for ASELSAN and MERL (Mitsubishi Electric Research Laboratories) during his graduate studies, where he conducted research in computer vision for surveillance.

Date: 29.03.2017 - 10:00-10:50

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Physics-Based Biped Motion Control in Computer Animation [24.03.2017 - 10:00-10:50]


Kinematic character animation techniques offer easy-to-implement and performance-efficient ways of virtually resembling human motions in computer animation applications. However, the resulting kinematic motions are limited to only pre-designed scenarios and completely unaware of the interactions with the environment. In order to achieve perfect realism, character animation research headed to physics-based approaches, which define the character with physical body-parts and maintain an active control of the character by applying joint torques.

In this talk, I will first give a short overview of our recent research studies on physics-based character animation. Afterwards, I will focus on our physics-based walking controller approach for synthesizing virtual character motions that closely mimic the styles of given reference motions. While the state of the art purely physics-based approaches enable creating physically realistic motions, their results are mostly robotic without a human-like style. With this motivation, our study focuses on delivering both physically and stylistically realistic results in a performance-efficient and easy-to-implement way. Specifically, we emphasize on the importance of tracking intra-step center of mass velocity variations for style resembling and devise a novel strategy with the purpose of better velocity tracking. We demonstrate the capabilities of our approach under various environment conditions with different reference motions obtained by either motion capture or manual authoring.

Zümra Kavafoğlu
Hacettepe University

Dr. Zümra Kavafoğlu is a research assistant at Department of Mathematics, Hacettepe University and received her B.Sc(2007), M.Sc.(2010) and Ph.D.(2016) degrees from the same department. During her M.Sc. studies, she performed research on computational geometry and scientific visualization as a part of the TÜBİTAK project “Development of Mine Automation and Design Software”. She conducted an interdisciplinary Ph.D. research that focused on physics-based biped motion controllers, as a part of the TÜBİTAK project “Data Driven Virtual Human Animation”. From December 2014 to June 2015, she was a visiting scholar in Utrecht University(the Netherlands), Department of Information and Computing Sciences with TÜBİTAK International Doctoral Research Fellowship. Her current research interests include physics-based and data-driven character animation, especially building general controllers with more efficient inverse dynamics approaches.

Date: 24.03.2017 - 10:00-10:50

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Search Result Diversification for the Web and Social Media [10.03.2017 - 11:00-11:30]


I will begin with a short potpourri of our most recent research, emphasis being on the topics related to the web search engines and social Web. Then, I will focus on the web search result diversification problem and present our novel contributions in two directions. First, I will introduce a set of new explicit diversification strategies based on the score(-based) and rank(-based) aggregation methods. Next, I will present how these diversification methods perform in the context of Tweet search, and how we improve them using word embeddings.

Ismail Sengor Altingovde

Ismail Sengor Altingovde is currently an associate professor in the Computer Engineering Department of Middle East Technical University (METU) at Ankara, Turkey. He has received his B.S., M.S. and Ph.D. degrees, all in Computer Science, from Bilkent University (Turkey). Before joining METU, he has worked as a postdoctoral researcher at Bilkent University and L3S Research Center in Hannover, Germany. He has worked in several national and international research projects, funded by the Scientific and Technological Research Council of Turkey and the European Union Sixth and Seventh Framework Programs. His research interests include web information retrieval with a particular focus on search efficiency, social web and web databases. He has published over 50 papers in prestigious journals and top-tier conferences. He is one of the recipients of Yahoo! Faculty Research and Engagement Program (FREP) award in 2013 and Young Scientist Award (GEBIP) of Turkish Science Academy (TUBA) in 2016.

Date: 10.03.2017 - 11:00-11:30

Location: Hacettepe University, Computer Engineering Building, Seminar Room

High Fidelity Sky Models [03.2017 - 10:00-11:00]


Light sources are an important part of physically-based rendering when accurate imagery is required. High-fidelity models of sky illumination are essential when virtual environments are illuminated via the sky as is commonplace in most outdoor scenarios. The complex nature of sky lighting makes it difficult to accurately model real life skies. The current solutions to sky illumination can be analytically based and are computationally expensive for complex models, or based on captured data. Such captured data is impractical to capture and difficult to use due to temporally inconsistencies in the captured content. This work enhances the state-of-the-art in sky lighting by addressing these problems via novel sky illumination methods that are accurate, practical and flexible. This work presents two novel sky illumination methods where; the first of which focuses on clear sky lighting and the second one deals with illumination from cloudy skies.

Pınar Satılmış (Hacettepe University)

Dr. Pınar Satılmış is a research assistant at the Department of Mathematics, Hacettepe University. She has a BSc with a first degree and a MSc from Department of Mathematics, Hacettepe University. She did her PhD in Visualisation Group (Warwick Manufacturing Group), University of Warwick, 2017. Her research interests are Machine Learning for Graphics Applications, Light Transport, Optimization Techniques, Image Processing, High Dynamic Range Imaging.

Date: 03.03.2017 - 10:00-11:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Modeling, Analysis and Autopilot Design for UAVs [24.02.2017 - 10:00-11:00]


In this talk, Dr. Coşku Kasnakoğlu will first give information about his Unmanned Aerial Vehicles (UAV) Laboratory and its research efforts. The laboratory conducts work on the modeling, analysis and autopilot design for UAVs. The systems built are verified using computer simulations, hardware-in-the-loop tests and actual flights. Future research directions including intelligent methods and fault-tolerant approaches will also be discussed.

Coşku Kasnakoğlu

Dr. Coşku Kasnakoğlu obtained B.S. degrees from the Department of Computer Engineering and the Department of Electrical and Electronics Engineering and at the Middle East Technical University (METU), Ankara, Turkey in 2000. He obtained his M.S. and Ph.D. degrees from the Department of Electrical and Computer Engineering at the Ohio State University (OSU), Columbus, Ohio, USA in 2003 and 2007. From 1996-2000 he worked as a researcher at TUBİTAK Information Technologies and Electronics Research Institute (BILTEN). From 2000-2007 he was as a graduate researcher at OSU. In 2008 he joined TOBB University of Economics and Technology, where he is currently an associate professor in the Department of Electrical and Electronics Engineering. Dr. Kasnakoglu's current research interests include dynamical modeling, analysis and autopilot design for aerospace systems with applications to unmanned aerial vehicles (UAVs). He is a member of IEEE, AIAA and has published numerous scientific papers in respected journals and conferences

Date: 24.02.2017 - 10:00-11:00

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Learning Actionable Representations from Large-Scale Visual Information [29.12.2016 - 10:30]


Humans created a tremendous value by collecting and organizing all their knowledge in publicly accessible forms, as in Wikipedia and YouTube. The availability of such large knowledge bases not only changed the way we learn, it also changed how we design artificial intelligence algorithms. Recently, propelled by the available data and expressive models, many successful computer vision and natural language processing algorithms have emerged. However, we did not see a similar shift in robotics. Our robots are still having trouble recognizing basic objects, detecting humans, and even performing simple tasks.
We believe the main limitation is, robots need to handle many challenges like scarcity of the supervision and the shift between different modalities and domains. We show that the common solution to all these problems is understanding the latent structure of the data. We propose machine learning algorithms, which can learn latent semantic structure with no supervision over multiple domains and modalities.

Ozan Sener
Stanford University

Ozan Sener is a PhD candidate at Stanford University co-advised by Ashutosh Saxena and Silvio Savarese. He earned his BS. and MS degrees from the Middle East Technical University. He later joined Cornell University as a PhD student. He later moved to Stanford University following his advisor's move. His research interests include machine learning, computer vision and robotics. His recent works involve transfer learning among domains and modalities, unsupervised learning using generative models, and exploring the problems relating functionality of objects to their shape.

Date: 29.12.2016 - 10:30

Location: Hacettepe University, Computer Engineering Building, Seminar Room

Entrepreneurship at Gaming Industry Seminar

IGDA Hacettepe Game Developers Academical Chapter represents "Oyun Sektöründe Girişimcilik" event which is the very first event of the GameDev101 series will be held on 22 December at 1:00 PM.

A. Emre Taş who is the CEO of the Alictus company will be the speaker. The purpose of the event is to gain experince about entrepreneurship at game development sectory for those plan to work on this sector or plan to start their own start-up.

Event link:

BBM601-701 Graduate Seminar Courses - Presentation Schedule
ACM Hacettepe Git WorkShop Event

Git Workshop will be held in Computer Engineering seminar room between 15:30 and 18:30 on Friday, December 9. Anyone interested is welcome to attend.

Git is the biggest version control and source code management system, used in software developing projects. Acting as a storage system containing whole history of a project's version control, Git also is an open source system.

Event link:

Presentations for introducing the research areas carried out in our department

Dear students,

In oderder to introduce the research areas carried out in our department, a series of presentations will be given by the faculty members on 9th and 16th of December. Full presentation schedule is provided below. Attandence is mandatory for all students who are currently registered at MSc. or PhD. programs and whose thesis supervisors have not been assigned yet. Undergraduate students who are planning to pursue graduate studies in the future are also strongly advised to attend these presentations.

December 2016
Saat Fri 9 Fri 16
09:30-09:45 Gonenc Ercan Suleyman Tosun
09:45-10:00 Burcu Can Mehmet Önder Efe
10:00-10:15 Harun ARTUNER Erkut Erdem
10:15-10:30 Kayhan İmre Aykut Erdem
10:30-10:45 Fuat Akal Nazli Ikizler
10:45-11:00 Ahmet Burak Can Pinar Duygulu
11:00-11:15 Adnan Ozsoy Murat Aydos
11:15-11:30 Ufuk Çelikcan Mehmet Koseoglu
11:30-11:45 Vahid Garousi Sevil Sen
11:45-12:00 Ayça Tarhan Lale Ozkahya
12:00-12:15 Fuat Akal

Game Development and Big Data Talks [18.11.2016 - 11:00]


11.00 - 12.30 Game Development with Unity - Hakan Sağlam - Seminar Hall

12.30 - 13.30 Lunch Break (sponsored by Peak Games)

14.00 - 16.00 Meet with Peak Games' Technical Leaders - Seminar Hall

6.30 - 17.30 A glance at the Big Data Infrastructure of Peak Games - Serdar Şahin - Yıldız Amphitheater M10

Date: 18.11.2016 - 11:00-17:30

Location: Seminar Hall, Computer Engineering Department / Yıldız Amphitheater

Identification of Legged Locomotion via Mechanics-Based and Data-Driven Models [11.11.2016 - 10:00]


The ultimate promise of legged locomotion on complex terrain led to both the construction of many legged robot platforms as well as mathematical models to describe their underlying dynamics. Motivated by the problems in the area, this study focuses on the identification of legged locomotor systems, in particular robotic legged locomotion via mechanics-based and data-driven models. In the first part, we worked on developing mechanics-based analytical models that can accurately represent center of mass trajectories of legged robots. We present systematic experiments on a well-instrumented monopedal robot platform to assess the model performance with physical robot data. On the other hand, representational power of such mechanics-based models are inevitably limited due to the nonlinear and complex nature of biological systems. To this end, we adopt a data-driven approach with the goal of furnishing an input—output representation of legged locomotor systems. Our method is based on formulating the identification problem of legged locomotion as a linear time-periodic system around a stable limit cycle and then develop frequency domain system identification techniques to obtain a state-space representation.

İsmail UYANIK, PhD
University of Bilkent

İsmail Uyanık is a Ph.D. student in Electrical and Electronics Engineering Department of Bilkent University. He received his B.Sc. and M.Sc. degrees from the same department in 2009 and 2011, respectively. His research has been devoted to discovering the principles by which animals move in nature. Inspired by biomechanical observations, he worked on developing novel system identification methods for identifying legged locomotion, in particular legged robotic systems. He also developed different legged robot platforms to experimentally validate his research findings. In 2014, he received Aselsan Ph.D. fellowship which is given to only a few Ph.D. students nationwide each year.

Date: 11.11.2016 - 10:00

Location: Seminer Salonu, Bilgisayar Mühendisliği Bölümü

Servise Yönelik Yazılım Mimarileri [09.11.2016 - 13:00]


Servise yönelik yaklaşım, servis tasarım ilkeleri, servise yönelik mimari standartları ve teknolojileri, İş süreci yönetimi ve servise yönelik mimari.

Yrd. Doç. Dr. Barış Özkan
Atilim Universitesi

Dr. Barış Özkan 10 yılı aşkın bir süredir yazılım ölçümü, yazılım maliyet modelleri, gereksinim mühendisliği, yazılım mühendisliği, iş süreçleri yönetimi, yazılım ürün ve hizmet kalitesi , BT yönetişimi, yönetimi ve denetimi konularında dersler vermekte ve araştırma çalışmaları yapmaktadır.

Date: 09.11.2016 - 13:00

Location: Seminar Hall, Computer Engineering Department

Graduate Research Workshop and Exhibition of Senior Projects [09.05.2016]

Action Recognition and Prediction with Applications to Daily Living [29.11.2015]


With the recent technological developments, egocentric cameras have become a part of our lives. They are designed as tiny cameras that can be worn without disturbing the person. In order to investigate the possible information gain from egocentric cameras, we used a multiple camera setting, where both an egocentric and multiple static cameras exist. Our research showed that when fused correctly, using the information from different types of cameras increases the recognition accuracy of actions. The model we proposed for this task is also suitable for other multi-modal settings. To prove the generality of the proposed model we also tested it on a setting with multiple static cameras and showed state-of-the-art results. Our model learns the importance of each camera in recognizing the actions, and it can also be used to direct the scenes automatically. We created examples of automatically directed scenes to show the concept.

We also addressed the problem of improving people's lives in a preventive way using egocentric cameras. In our work preventive refers to the general notion of reminders that can possibly prevent people from making mistakes that can cause problems. For example, when people are leaving a room while the stove is on, they might be reminded to turn the stove off. We proposed a notification decision mechanism that reasons about interdependencies between actions, checks at every time step whether there is a missing action that should be completed before the ongoing one ends, calculates a cost for missing it, and uses this cost to make a notification decision. Such a notification system requires recognizing the past actions and predicting the ongoing action while segmenting the activity observed so far. For this purpose, we proposed a model that uses standard features and accomplishes these three tasks successfully. We showed promising results on the extremely challenging task of issuing correct and timely reminders on a new egocentric dataset.

Bilge Soran, PhD
University of Washington

Bilge Soran is a Post-Doctoral researcher at the University of Washington, Mechanical Engineering Department working on healthcare applications of computer vision. She earned a Ph.D. degree from the University of Washington, Computer Science and Engineering Department in 2015, under the supervision of Linda Shapiro and Ali Farhadi. Her thesis topic was "Action Recognition and Prediction with Applications to Medical Diagnosis and Daily Living". She holds a M.Sc. degree from the same department, where the thesis topic was Parcellation of Human Inferior Parietal Lobule Based on Diffusion MRI. She earned another M.Sc. degree from the Gebze Institute of Technology, in 2007, where the thesis topic was Event Driven Molecular Dynamics Simulation. Before that, she obtained her B.Sc. degree from the Middle East Technical University, in 2002. She worked in The Scientific and Technological Research Council of Turkey between 2003-2009, as a researcher and senior researcher on avionics and simulation systems. Before that she had 2 years of industry experience.

Action Perception in the Human Brain [11.11.2015]

Successfully perceiving and recognizing the actions of others is of utmost importance for the survival of many species. For humans, action perception is considered to support important higher order social skills, such as communication, intention understanding and empathy, some of which may be uniquely human. Over the last two decades, neurophysiological and neuroimaging studies in primates have identified a network of brain regions in occipito-temporal, parietal and premotor cortex that are associated with perception of actions, also known as the Action Observation Network. Despite growing body of literature, the functional properties and connectivity patterns of this network remain largely unknown.

One of the goals of my research is to address these general questions about functional properties and connectivity patterns with a specific focus on whether this system shows specificity for biological agents. To this end, we collaborated with a robotics lab, and manipulated the humanlikeness of agents that perform recognizable actions by varying visual appearance and movement kinematics. We then used a range of measurement modalities including cortical EEG oscillations, event-related brain potentials (ERPs), and functional magnetic resonance imaging (fMRI) together with a range of analytical techniques including pattern classification, representational similarity analysis (RSA), and dynamical causal modeling (DCM) to study the functional properties, temporal dynamics, and connectivity patterns of the Action Observation Network.

While our findings shed light whether the human brain shows specificity for biological agents, the interdisciplinary work with robotics also allowed us to address questions regarding human factors in artificial agent design in social robotics and human-robot interaction such as uncanny valley, which is concerned with what kind of robots we should design so that humans can easily accept them as social partners.

Burcu Aysen Urgen received her PhD (2015) in Cognitive Science from University of California, San Diego (UCSD) under the supervision of Professor Ayse P. Saygin; her BS degree in Computer Engineering and Information Science from Bilkent University, and her MS degree in Cognitive Science from Middle East Technical University. She is currently a post-doc in University of Parma, Italy with Professor Guy Orban. Her primary research interest is the neural mechanisms underlying visual perception of actions in the human brain. In her PhD, she investigated whether the human brain shows specificity for biological agents using humanoid robots by collaborating with Professor Hiroshi Ishiguro's lab (Osaka University, Japan), neuroimaging methods including fMRI and EEG, and machine learning techniques. Her interdisciplinary work has been published and presented at international and interdisciplinary journals and conferences. She was also one of the three awardees of the Interdisciplinary Scholar Award by UCSD with her interdisciplinary work between cognitive neuroscience and human-robot interaction.

-- Burcu Aysen Urgen, PhD

Quantum Cuts : A Quantum Mechanical Spectral Graph Partitioning Method for Salient Object Detection, Çağlar Aytekin, Tampere University. [06.10.2016]

Çağlar Aytekin, Department of Signal Processing, Tampere University of Technology

Title: Quantum Cuts : A Quantum Mechanical Spectral Graph Partitioning Method for Salient Object Detection

Date: October 6, 2016 Thursday

Place: Seminar Hall, Computer Engineering Department

Time: 15:00

Abstract: Salient Object Detection can be defined as highlighting the objects that visually stand out from the rest of the environment. The growing interest in this topic can be explained by its many direct applications such as image and video coding, summarization and object tracking. Moreover, it can be used to reduce the complexity of higher level Computer Vision tasks by reducing their search space. This talk focuses on this important and emerging Computer Vision task and describes novel approaches and methods for salient object detection in images and videos.
First part of the talk is constituted of a direct application of ideas originally proposed for describing the wave nature of particles in Quantum Mechanics and expressed through Schrödinger's Equation, to salient object detection in images. Second part introduces a spectral graph based salient object detection method, namely Quantum Cuts. The proposed method adopts a novel approach to spectral graph partitioning by integrating quantum mechanical concepts to Spectral Graph Theory. The third part of the talk covers improvements and extensions on Quantum Cuts by analyzing the main factors that affect its performance in salient object detection. Finally future directions in salient object detection research will be discussed.

Bio: Çağlar Aytekin was born in Turkey in 1987. He received the B.Sc. degree from the Electrical and Electronics Engineering Department from Middle East Technical University, Ankara Turkey in 2008. He received the M.Sc. department with Signal Processing major from Middle East Technical University, Ankara, Turkey in 2011. He is currently a Ph.D student and researcher in Department of Signal Processing, Tampere University of Technology. He has published 16 papers in international conferences and journals. He has won the the ICPR 2014 Best Student Paper Award for his work in unsupervised salient object detection. His research interests are, visual saliency, semantic segmentation, quantum mechanics based computer vision methods and deep learning.

Human activity analysis, Selen Pehlivan, TED University [09.06.2016]

Selen Pehlivan, TED University

Title: Human activity analysis

Date: June 9, 2016 Thursday


Abstract: In this talk, I will summarize our work on human activity analysis. First, focusing on multiple camera views, I will describe our method for fusion of frame judgments. We have shown that when there are enough overlapping views to generate a volumetric reconstruction, our recognition performance is comparable with that produced by volumetric reconstructions. Then, I will shortly introduce our recent work on weakly-supervised spatial-temporal action localization.

In the second part of the talk, I will present our work on fine-grained categorization. We introduce a unified approach for classification and keypoint localization, targeting subcategories with wide pose variations and high intra-class similarities. We apply our method to identifying instances of birds.

Finally, I will present our recent effort on action understanding to investigate the representational properties of the Action Observation Network using fMRI and modeling. In our study, we used powerful computational tools from computer vision as well as attribute-based semantic models to represent the videos consisting of natural actions, and linked those models to the brain responses (fMRI) using representational similarity analysis.

Bio: Selen Pehlivan holds a B.S. (Bilkent University, 2004) and an M.S. (Koç University, 2006) in Computer Engineering. Between 2004 and 2006, she was a member of the 3DTV EU Project. She got her Ph.D. degree (Bilkent University, 2012) also in Computer Engineering. From August 2009 to February 2011, she was a visiting scholar in Department of Computer Science at University of Illinois at Urbana-Champaign (UIUC). During her graduate studies, she worked as a research and teaching assistant. Before joining TED University, she was a postdoctoral research associate in Computer Vision Group at University of Central Florida (UCF).

Dr. Pehlivan is working on computer vision and machine learning with a primary focus on human activity understanding with video interpretation and object recognition.

Inter Machines Communication [3.4.2015]

Life Sciences Data Management in a Research Center [27.3.2015]

Ultimate Trends in Software Engineering [19.3.2015]

Speech Recognition Systems [13.3.2015]

Forensic and Latest Topics [6.3.2015]

Big Data Mangement with Oracle [26.12.2014]

On Chip Networks [18.12.2014]

Problems in extremal graph theory[11.12.2014]

Management [3.12.2014]

R&D and Technology in Defense Industry [28.11.2014]

Smart Energy [21.11.2014]

Digital Companies and Excellence in business process [14.11.2014]

Treats and Defense in Network Security [7.11.2014]

Wireless Sensor Networks [24.10.2014]

Hacettepe University Department of Computer Engineering
06800 Beytepe Ankara