Ozan Karaali

• Wrote a thesis titled ”Learning-Based Ultrasound Image Super-Resolution”. • Developed and evaluated two proposed deep learning SR methods for ultrasound images on Pytorch, benchmarked against various state-of-the-art approaches. • The first method integrates a Swin Transformer into Deblurgan v2’s feature pyramid network for enhanced deblurring and super-resolution. The second method employs a HAT-NAF mixture model to optimize the Hybrid Attention Transformer network further.

• Developed a dynamic business rule editor that generates the rules for consistency checker, which in turn validates the graph data of the relation maps of building data. • Maintained the consistency checker of the metis-II project by documenting, fixing the bugs, refactoring the code with OOP practices and also implementing CI/CD pipelines to ensure code sustainability and maintainability.

Project EVOLVE: Pioneered the parallelization of optimization algorithms for ride-hailing, deployed on an HPC center with Python/OpenMPI, gauging significant performance benefits. Project GeoCoding QA: Evaluated latitude/longitude resolution of providers using metrics like Levenshtein distance and Haversine distance. Project Powerpool: Identified error causes in BMW's energy market backend for EV charging, implemented unit tests to prevent a recurrence, and developed a tool to speed up error analysis.

• Developed the interdisciplinary project ”Deep Learning in the Architectural Design Process”, under the metis-II project. • This project aims to build an LSTM and FCNN network mixture that can sense time and featurewise changes and predict the architect’s next design intentions by feeding the drawing data in time series. Which is used in the metis-II AI framework for intelligent auto-completion of building designs.

• Learned the basics of a neural network and trained models to classify images using Caffe in a Docker environment using multiple GPUs. Processed video frames using the GStreamer filter plugin, which I developed with a Python script of OpenCV bridged via Boost::Python and refined the GUI application as a continuation of my summer internship.

• Developed IOS applications using Swift and RxSwift with Adesso Mobile Bootcamp program. • Learned the differences between Android’s development environment (prior knowledge from the Mobile Applications course) and iOS’s environment, such as Storyboard, UI, Pods, CoreData, Alamofire, etc., and wrote unit tests of the application. • Refactored an app using MVC architecture to MVVM architecture to better understand the Reactive programming paradigm when propagating data inside the application using RxSwift.

• Developed a GUI application to be used for video/image processing purposes using Python/QT, and also started to learn how video processing works frame by frame using GStreamer and OpenCV.

• Developed an internal tool as a web application for analysis of traffic data collected from Bluetooth using Flask, Pandas with Python, and React. Learned basics of web development, frontend design, REST architecture, backend, and some basic data science algorithms and their applications.

• Area of Specialization: Computer Vision — Supplemental Areas: Machine Learning, Distributed Systems