My Projects

• Developed a chess playing robot arm utilizing Raspberry Pi 3B and 3D printed parts for hardware.

• Used SORNet (Spatial Object-Centric Representation Network) algorithm to detect chess pieces in one-shot manner.

• Developed a face application for the robot arm to show the face expressions of the robot. Pre-configured emotions and expressions for robot arm to execute during playing chess with the user.

• Developed a system for training various RL (Reinforcement Learning) algorithms, optimizing hyper-parameter values and comparing the results.

• Using the developed system, trained an innovative RL algorithm (Parallel Experience Transfer Multi Agent RL) for driving a self-driving car.

• The trained algorithm achieved the third place in the self-driving car competition sponsored by Eatron Technologies.

• Predicted customer churn in telecommunication companies with machine learning techniques.

• Used explainable AI to analyze the prediction results and delivered a detailed report for the churning reasons

• Used Invarian Information Clustering method on the component image dataset (FICS-PCB) of printed circuit boards (PCB).

• Used self-learning to calculate mutual information and optimized the network.

• Used pseudo-labeling to find labels of the corresponding clusters.

• Analyzed and compared the results with previous studies.

• Researched state-of-the-art methods for unsupervised image clustering (IIC, Autoencoders, Deep Clustering, etc.) and evaluated the results with some of the knwon image datasets such as CIFAR-10, CIFAR-100, STL-10

• Reported the comparison of the methods

• Compared the results with classic approach of feature extraction on images + K-means or KNN

• Implemented different versions of VGGNet and ResNet on Keras. Evaluated image dataset of an active kaggle competition n leaf diseases.

• Tried various combinations of transfer learning, batch normalization, data augmentation, and completed the hyperparameter optimization with grid search.

• Used Extended kalman filter (EKF) and particle filter to localize a wheeled ground robot.

• Used EKF with SLAM (simultaneous localization and mapping) algorithm to generate the map of the environment in ROS with Gazebo and RViz.

Erasmus+ International Collaboration Project

• Currently working with a team of four international people to make a market analysis and achieve a technique to improve energy consumption on base stations of 5G networks.

• Using Matlab and NS3 simulations to optimize the hyper parameters of the algorithm using reinforcement learning. Using various data mining algorithms to analyze and visualize for a case study.

• It will be possibly extended as a reinforcement learning project which uses the simulation we completed as the environment.

for Tübitak Efficiency Challenge - Autonomous Category

• Currently implementing a computer vision application to detect traffic elements in real time using the YOLO object detection system.

• Working with a team of four to enable the project to work on a Gazebo simulation and the ROS.

• Worked remotely (with Git and Microsoft Teams) on a sharing economy project in an R&D team (confidential).

• Made a mobile application which consists of renting, and navigating electric vehicles, with features such as feedback, onboarding screen, authentication, qr scanning, bluetooth connection, map API, and a credit system.

Technologies using: Java, Firebase, MapBox API, Scrum, Android Studio

• Took part in the development and testing of building a REST API for database querying in the backend.

• Used various frameworks such as Node.js, Ramda.js, and Mocha.js.

• Carried out agile teamwork with a team of six international people in Scrum methodology.

• Used project management tools like Github, Trello, and Slack efficiently during the progress of the project.

Worked on a project which aims to predict emotions in real-time by combining facial expressions and speech features (acoustic and linguistic).

• Achieved better accuracy results than the current state of the art model. (CNN with SVM and ResNet50)

• Used different pre-processing techniques such as face frontalization, detection, and kirsch compass masks for various datasets. (FER2013, JAFFEE, CK+ and IEMOCAP).

• Technologies used: Python, Tensorflow, Keras, OpenCV, Dlib, LibROSA, Scrum, Git

Analysis of Algorithms Course Project

• Implemented the K-means clustering algorithm and the K-NN classification algorithm in both Python (Scikit-learn) and C programming languages.

• Compared running times and memory usages between these implementations using various datasets.

Artificial Intelligence Course Project

• Used Simulation of Urban Mobility framework, and Keras module with Tensorflow backend in Python.

• Applied Machine Learning, Deep Reinforcement Learning algorithms (Deep Q-Learning, Proximal and Trust Region Policy Optimization) to optimize signalization in various traffic environments.

Database Management Course Project

• Made a Hotel Management Application on Java using Swing framework.

• Created a database with MySQL and integrated with the desktop application.

Mobile Programming Course Project

Provides statistics and gives notifications about the users and applications of the Steam platform
• Android’s advanced components such as notifications, services, broadcast and content providers.
• Synchronized SQLite database in Android app and used different web APIs used to retrieve data.

Object-Oriented Programming Course Project

• 2D side-scrolling multiplayer computer game developed with Java programming language.