Explainable AI & Uncertainty Quantification

During my ERCIM postdoc fellowship at Fraunhofer IEM, I worked on model agnostic explainability methods for post hoc explanations of machine learning models, especially time series prediction models. I mostly focusted on counterfactual explanations.

I was part of the team which was responsible for developing the search functionalities of Huawei App Gallery where we collaborated with Huawei HQ. I led the Multilingual Search & Ranking Research Team in Turkey and we were responsible for Africa, Russia, Latin America, and Asia regions.

Video Advertisement Placement

This was an industrial research project at Huawei and it was partially funded by The Scientific and Technological Research Council of Turkey (TUBITAK). The aim in this project is to place advertisements in videos by selecting an empty area, select an ad from the ads database that is related to the contents of the scene, and place it by 3D transformations and handling occlusions. We aim to show ads without disturbing the viewers. This can be thought as a purely vision based, markerless augmented reality application. Researh efforts in this project span a wide range of topics like video content understanding and geometric scene analysis, local feature extraction and tracking, and some basic natural language processing and audio signal analysis. I was the project manager in this project and I took part in all phases including idea development, grant application, and I lead the research efforts. Here’s our patent application: Device and method for processing a digital video. TUBITAK support duration was 2 years, and the budget (including the partial in-house funding by Huawei) was a little above TRY 2M (around USD 530K at that time).

Hyperspectral Image Classification

My PhD thesis and a research project at Gebze Technical University led by Prof. Koray Kayabol and funded by TUBITAK.

Video Content Modeling and Content Based Copy Detection

This was a research project at ITU MSPR led by Prof. Bilge Günsel and funded by TUBITAK. We attended TRECVID several years in the content based copy detection track.

Content Based Image Search Using Local Features

A project at ITU MSPR. It was funded by Avea (now Turk Telekom).

Content Based Image & Video Search for Enterprise Databases

An industrial research project at Huawei.

User Modeling and Intelligent Campaign System for Mobile Operators

An industrial research project at Huawei.

Tissue Density Classification in Mammographic Images

MSc thesis at Istanbul Technical University.

High Precision Location Improvements for a GNSS Receiver

This was a project at Netas. I worked on the location corrections by using the satellite augmentation signals. This included decoding the signal to obtain the correction messages and applying the corrections to the locations calculated by GPS (and similar satellite systems). I implemented on an embedded DSP system.