Bilgiç, Emirhan; Sengör, Neslihan Serap; Yalabik, Namik Berk; Isler, Yavuz Selim; Gelen, Aykut Gorkem; Elibol, Rahmi

This study examines the integration of Contrastive Predictive Coding (CPC) with Spiking Neural Networks (SNN). While CPC learns the predictive structure of data to generate meaningful representations, SNN mimics the computational processes of biological neural systems over time. In this study, the goal is to develop a predictive coding model with greater biological plausibility by processing inputs and outputs in a spike-based system. The proposed model was tested on the MNIST dataset and achieved a high classification rate in distinguishing positive sequential samples from non-sequential negative samples. The study demonstrates that CPC can be effectively combined with SNN, showing that an SNN trained for classification tasks can also function as an encoding mechanism. Project codes and detailed results can be accessed on our GitHub page: https://github.com/vnd-ogrenme/ongorusel-kodlama/tree/main/CPC_SNN

Gelen, Aykut Gorkem; Sengor, Neslihan Serap; Elibol, Rahmi; Isler, Yavuz Selim; Bilgiç, Emirhan; Yalabik, Namik Berk

Pattern recognition problems can be solved with spiking neuron models. In this study, the pattern recognition performance of an architecture built with groups that have excitatory and inhibitory neural dynamics was investigated. Examples from the MNIST dataset were converted into spike signals and used as stimuli. The goal of the research is to create a clustering model based on spiking neurons and train it using unsupervised learning. In the study, connections between neuron groups were designed to create competition, and constraint schemes based on the winner-take-all mechanism were used. Because the labels are unknown, accuracy values cannot be calculated like in classification problems. As a result, intra-group accuracy percentages were calculated, and the accuracy of the architecture was evaluated. As a result, the results of this research show that models based on spiking neuron groups can successfully solve unsupervised pattern recognition problems.

Tas, Mehmet Bilge Han; Özkan, Kemal; Sariçiçek, Inci; Yazici, Ahmet

With the increase in population and transportation vehicles, increasing transportation times due to traffic density and carbon dioxide emitted by vehicles is one of the most important problems today. In this study, it is aimed to reduce these times and reduce traffic density by controlling traffic lights within the scope of smart cities. By using Policy Gradient-based reinforcement learning, a realistic city traffic sample is created with various vehicle elements such as cars, motorcycles and bicycles, and traffic lights are controlled to provide a smoother traffic with lower transportation times. Thanks to the application, a one-third improvement was achieved in the 180-hour period presented as an example.

Aslan Yildiz, Özge; Sariçi̇Çek, İnci; Özkan, Kemal; Yazici, Ahmet

Intelligent transportation systems are an important issue within the scope of smart cities. Vehicle routing problems, which are a combinatorial problem, need to be solved in the design of the relevant systems. In this regard, the use of artificial intelligence optimization algorithms such as meta-heuristics has increased significantly in recent years. To the best of our knowledge, there are no papers that address the Dial and Ride problem by using reinforcement learning, one of the learning-based models. In this study, the Dial and Ride problem is solved for a single service vehicle using the transformer-based deep reinforcement learning method. The proposed method is tested on a problem generated in an environment in Eskişehir Büyükdere Neighborhood. As a result of the test problem, it is shown that the proposed method produced a solution to the problem in a reasonable time. The study showed that the Dial and Ride Problems can be solved with reinforcement learning.

Elibol, Rahmi

The spines on the dendrites of neurons have many biophysical functions such as increasing the surface area of the dendrites and increasing the input resistance of neuron. It is known that synaptic plasticity is effective in the formation and growth of spines. In this work, a computational model of growth of dendritic spines due to synaptic plasticity is presented. The proposed model is built with spiking neural networks and synaptic dynamics. In the model, the change of synaptic weights is achieved with synaptic plasticity, and the growth of dendritic spines can be demonstrated by using this change.

Yildiz, Eyyup; Ensari, Tolga; Sener, Leyla Turker

Flow Cytometry device is frequently used in the analysis of blood samples. Analysis of Flow Cytometry data is needed in cases such as diagnosing disease, monitoring progression of disease. However, the manual analysis of these multi-dimensional data by the hand cannot be performed at the desired level due to various reasons. In this study, it is aimed to create an algorithm which allows the gating process by manual gating to be performed automatically by the experts in the flow cytometry data. The algorithm consists of k-means, Gaussian Mixture Method (GMM) clustering methods, Expectation Maximization (EM) algorithm and Chernoff distance measurement method. The algorithm developed in the scope of the study was tested on the DLBCL dataset and a success rate of 86.06% was obtained.

Aslan, Ozge; Yazici, Ahmet

This study deals with the problem of conflict-free route planning for multiple autonomous transport vehicles. In this study, an algorithm based on A* algorithm to solve conflict-free routing problems for multiple autonomous transport vehicles, is proposed. While classical A* algorithm plans routes for a single start and end node at the same time, the proposed algorithm plans routes for multiple start and end nodes without conflicts. The proposed algorithm apply waiting or change the route strategies to resolve conflict types. The proposed algorithm has been tested on multiple autonomous transport vehicles having different velocities (heterogeneous). Tests conducted within the scope of the study, show that the proposed algorithm finds effective conflict-free routes.

Cabuk, Umut Can; Akram, Vahid Khalilpour; Dagdeviren, Orhan

Wireless sensor networks (WSN) and their applications become more and more common as the technology advances. The emerging concepts of Internet of Things and the Industry 4.0 boost their popularity, too. But there are still dominant bottlenecks including battery lifetime, transmission range, routing issues and others. These limitations may cause permanent faults on nodes, and thus, cause exclusion of the nodes from the network, which eventually may disconnect the network. Minimum cut size, a metric regarding the edge-connectivity of a graph, can be used to measure the reliability of a WSN. This work discusses that role of the minimum cut size; and more importantly, presents the results of a fundamental simulation showing the correlations between the total node count, the transmission range, and the achieved minimum cut size as an indicator of the maximum flow, in a randomly generated WSN topology. Our simulation results showed that at least 100 nodes with a transmission range of 140 m are required to create a random connected network on a field of 1000× 1000 m area. If the ranges of the nodes be 80 m, then at least 200 nodes should be distributed on the field to expect a reliable connected network.

Aydin, Yildiz; Dizdaroglu, Bekir

Image inpainting is a reconfigurable reconstruction of detected (marked) areas. This study focuses on damaged images and key point-based attributes such as speeded-up robust features (SURF), scale invariant feature transformation (SIFT), Harris algorithm and features of maximally stable extremal regions (MSER) are used for this purpose. It is assumed that there are images, one of them is to be inpainted and another of them is relevant to be taken from different viewpoints or from different geometric transformation, and the Affine transformation differences between the image to be inpainted and the relevant image are determined. In line with the information obtained, these two images are restructured to be at the same scale and in the same position. Experimental results show that the damaged image is successfully inpainted depending on the images of different scales and angles.

Aydin, Yildiz; Muzaffer, Gul; Ulutas, Guzin

Digital images have become questionable as to how reliable they are with the increased use of the field and the importance of it. Especially since digital images are important in such important areas, it becomes questionable how reliable these images are. One of the types of digital image forgery that is easily done with the use of image editing software is copy move forgery. In this study, we propose a new keypoint-based method to detect copy move forgery. The proposed method was compared with the Speeded-Up Robust Features (SURF) algorithm based study in the literature and demonstrated performance advantage even under some attacks.