AKAhmet Bugra Koku

Ahmet Bugra Koku

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Ahmet Bugra Koku has a total of 79 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Training-Free Affordance Labeling and Exploration Using Subspace Projection and Manifold Curvature Over Pre-Trained Deep Networks

    by İsMaıl Özçıl, A. Buğra Koku

    The advancement in computing power has significantly reduced the training times for deep learning, enabling the rapid development of networks designed for object recognition. However, the exploration of object utility, the object’s affordance, as opposed to object recognition, has received comparatively less attention. Existing object affordance models exhibit shortcomings, including limited robustness across diverse architectures and insufficient performance in complex environments. This work focuses on using pre-trained networks trained on object classification datasets to explore object affordances. While these networks have proven instrumental in transfer learning for classification tasks, the presented approach in this study diverges from conventional object classification methods by labeling affordances without modifying the final layers. Instead, pre-trained networks are employed to learn affordance labels without requiring specialized classification layers. Two approaches are tested: the Subspace Projection Method and the Manifold Curvature Method, which facilitate the determination of affordance labels without such modifications. Both the Subspace Projection Method and the Manifold Curvature Method were evaluated using nine distinct pre-trained networks across two different affordance datasets. The Subspace Projection Method achieved a True Positive Rate of up to 94% and 96% for the best-performing networks on each dataset, while the Manifold Curvature Method attained True Positive Rates exceeding 98% and 99% with its top-performing networks. Furthermore, both methods identify affordance labels that are not marked in the ground truth but are present in various cases. The robustness of the Manifold Curvature Method and the exploration capability of both methods highlight the effectiveness of proposed techniques for affordance labeling.

  • Open Access

    Affordance Labeling and Exploration: A Manifold-Based Approach

    by İsmail Özçil, A. Buğra Koku

    The advancement in computing power has significantly reduced the training times for deep learning, fostering the rapid development of networks designed for object recognition. However, the exploration of object utility, which is the affordance of the object, as opposed to object recognition, has received comparatively less attention. This work focuses on the problem of exploration of object affordances using existing networks trained on the object classification dataset. While pre-trained networks have proven to be instrumental in transfer learning for classification tasks, this work diverges from conventional object classification methods. Instead, it employs pre-trained networks to discern affordance labels without the need for specialized layers, abstaining from modifying the final layers through the addition of classification layers. To facilitate the determination of affordance labels without such modifications, two approaches, i.e. subspace clustering and manifold curvature methods are tested. These methods offer a distinct perspective on affordance label recognition. Especially, manifold curvature method has been successfully tested with nine distinct pre-trained networks, each achieving an accuracy exceeding 95%. Moreover, it is observed that manifold curvature and subspace clustering methods explore affordance labels that are not marked in the ground truth, but object affords in various cases.

  • Open Access

    Mathematical and Machine Learning Approaches for Classification of Protein Secondary Structure Elements from Cα Coordinates

    by Ali Sekmen, Kamal Al Nasr, Bahadir Bilgin, Ahmet Bugra Koku, Christopher Jones

    Determining Secondary Structure Elements (SSEs) for any protein is crucial as an intermediate step for experimental tertiary structure determination. SSEs are identified using popular tools such as DSSP and STRIDE. These tools use atomic information to locate hydrogen bonds to identify SSEs. When some spatial atomic details are missing, locating SSEs becomes a hinder. To address the problem, when some atomic information is missing, three approaches for classifying SSE types using Cα atoms in protein chains were developed: (1) a mathematical approach, (2) a deep learning approach, and (3) an ensemble of five machine learning models. The proposed methods were compared against each other and with a state-of-the-art approach, PCASSO.


  • Open Access

    Robust feature space separation for deep convolutional neural network training

    by Ali Sekmen, Mustafa Parlaktuna, Ayad Abdul-Malek, Erdem Erdemir, Ahmet Bugra Koku

    This paper introduces two deep convolutional neural network training techniques that lead to more robust feature subspace separation in comparison to traditional training. Assume that dataset hasMlabels. The first method createsMdeep convolutional neural networks called $$\{\text {DCNN}_i\}_{i=1}^{M}$${DCNNi}i=1M . Each of the networks $$\text {DCNN}_i$$DCNNi is composed of a convolutional neural network ( $$\text {CNN}_i$$CNNi ) and a fully connected neural network ( $$\text {FCNN}_i$$FCNNi ). In training, a set of projection matrices $$\{\mathbf {P}_i\}_{i=1}^M$${Pi}i=1M are created and adaptively updated as representations for feature subspaces $$\{\mathcal {S}_i\}_{i=1}^M$${Si}i=1M . A rejection value is computed for each training based on its projections on feature subspaces. Each $$\text {FCNN}_i$$FCNNi acts as a binary classifier with a cost function whose main parameter is rejection values. A threshold value $$t_i$$ti is determined for $$i^{th}$$ith network $$\text {DCNN}_i$$DCNNi . A testing strategy utilizing $$\{t_i\}_{i=1}^M$${ti}i=1M is also introduced. The second method creates a single DCNN and it computes a cost function whose parameters depend on subspace separations using the geodesic distance on the Grasmannian manifold of subspaces $$\mathcal {S}_i$$Si and the sum of all remaining subspaces $$\{\mathcal {S}_j\}_{j=1,j\ne i}^M$${Sj}j=1,j≠iM . The proposed methods are tested using multiple network topologies. It is shown that while the first method works better for smaller networks, the second method performs better for complex architectures.


  • Senior Project Design Methodology

    by Ahmet Bugra Koku, Ali Sekmen, Ismail Fidan

    Senior and Capstone Design Project courses are widely adapted in engineering and computer science curricula to prepare students for the industry and give them the opportunity to work in real-world projects. One of the motivations behind offering such courses is to satisfy the ABET criteria emphasizing the importance of functioning in multi-disciplinary teams. Senior project courses in computer science/engineering are usually designed to be one or two semester-long and they traditionally involve large-scale software systems development. There is also a tendency to assign hardware/software integration projects to prepare students for the competitive job market. This paper introduces an innovative approach to enable students to work in multi-disciplinary teams and develop large-scale software/hardware systems. Programming language and operating system independent component-based software development is extensively discussed and explained with some complex senior project examples.

  • Fast head detection in arbitrary poses using depth information

    by Akif Hacinecipoglu, Erhan Ilhan Konukseven, Ahmet Bugra Koku

    Purpose: This study aims to develop a real-time algorithm, which can detect people even in arbitrary poses. To cover poor and changing light conditions, it does not rely on color information. The developed method is expected to run on computers with low computational resources so that it can be deployed on autonomous mobile robots.Design/methodology/approach: The method is designed to have a people detection pipeline with a series of operations. Efficient point cloud processing steps with a novel head extraction operation provide possible head clusters in the scene. Classification of these clusters using support vector machines results in high speed and robust people detector.Findings: The method is implemented on an autonomous mobile robot and results show that it can detect people with a frame rate of 28 Hz and equal error rate of 92 per cent. Also, in various non-standard poses, the detector is still able to classify people effectively.Research limitations/implications: The main limitation would be for point clouds similar to head shape causing false positives and disruptive accessories (like large hats) causing false negatives. Still, these can be overcome with sufficient training samples.Practical implications: The method can be used in industrial and social mobile applications because of its robustness, low resource needs and low power consumption.Originality/value: The paper introduces a novel and efficient technique to detect people in arbitrary poses, with poor light conditions and low computational resources. Solving all these problems in a single and lightweight method makes the study fulfill an important need for collaborative and autonomous mobile robots.


  • Open Access

    Slippage Estimation of Two Wheeled Mobile Robot Using Recurrent Deep Neural Network

    by İsMail Özçil, A. Buğra Koku, E. İlHan Konukseven

    Position, velocity and acceleration information are important for mobile robots. Due to the wheel slippages, encoder data may not be reliable and IMU data also contains a cumulative error. Errors of inertial measurements are accumulated over velocity and position estimates and as time increases, these errors grow higher. Due to robot hardware and the operating surface, ground truth may not be available. In this work recurrent deep neural network is proposed in order to reduce the error in speed and yaw angle estimates coming from encoder and IMU data. Neural networks are commonly used to capture the behavior of linear and nonlinear systems. Since ground-wheel interaction forces are modeled with non-linear models such as the Magic formula and determining parameters of those models require time and test setups, there is a need for simpler methods to model the behavior of simple robots. Neural networks could be used to model non-linear systems. In this work, a recurrent deep neural network is proposed to estimate the speed and yaw angle of a two-wheeled differentially driven mobile robot. Using the information coming from the camera positioned above the test area as ground truth, the network is trained. After that, the output of the network is recorded in the absence of ground truth information in the network. Finally, the performance of the network is evaluated using network output, sensor data calculation, and ground truth.