Ozgur Kara

Ozgur Kara

Incoming Machine Learning PhD @ GAtech

Bogazici University


I am currently a senior Electrical Electronics Engineering student at Bogazici University, Turkey. I am pursuing Machine Learning PhD at Georgia Institute of Technology starting Fall'22.

The ultimate goal of my research is threefold. First, I want to delve deeper into the newly developing machine learning paradigms such as continual learning, few-shot learning, meta-learning. Second, I would like to build explainable and controllable systems, with a specific interest in GANs. Third, combining these, I would like to build robust machine learning systems for computer vision applications (e.g. medical imaging, video understanding).

Throughout my research life, I have collaborated with top-notch research labs throughout the world: Affective Intelligence & Robotics Lab - University of Cambridge, Explainable Machine Learning Group - University of Tuebingen, Neuro Intelligent Systems Lab - Emory University, Nanonetworking Research Group - Bogazici University, Computer Vision Lab - ETH Zurich.

Download my resumé.

  • Deep Learning
  • Computer Vision
  • Meta Learning
  • Machine Learning PhD, Incoming (2022)

    Georgia Institute of Technology

  • BSc in Electrical-Electronics Engineering, 2018 - 2022 (Expected)

    Bogazici University

  • Math & Science, 2013 - 2018

    Kadikoy Anadolu High School

Recent News

All news»

[22/04/2022] I will be interning at VILAB-EPFL under Dr. Amir Zamir for 2022 Summer.

[09/04/2022] I have been accepted at the Machine Learning PhD program at Georgia Tech, starting Fall'22.

[02/03/2022] First “First Authorship” paper on computer vision was accepted to CVPR 2022.

Publications & Submissions

(2022). ISNAS-DIP: Image-Specific Neural Architecture Search for Deep Image Prior. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).

PDF Code Dataset

(2021). Domain-Incremental Continual Learning for Mitigating Bias in Facial Expression and Action Unit Recognition. Under Review - IEEE Transactions on Affective Computing.


(2021). Molecular Index Modulation using Convolutional Neural Networks. Under Review - Elsevier Nano Communication Networks Journal.

PDF Code

(2021). Towards Fair Affective Robotics: Continual Learning for Mitigating Bias in Facial Expression and Action Unit Recognition. In Workshop on Lifelong Learning and Personalization in Long-Term Human-Robot Interaction, 16th ACM/IEEE International Conference on Human-Robot Interaction.

PDF Video


Bogazici University - Creative AI Technologies Research Lab
Undergraduate Researcher
Sep 2021 – Present Istanbul
Doing research on the discovery of controllable GANs and latent space manipulation with the aim of building controllable GANs.
ETH Zurich - Computer Vision Lab
Undergraduate Researcher
Jan 2021 – Mar 2021 Zurich (Remote)
Worked on Deep Image Prior (DIP), an unsupervised approach for image restoration task, and propose to apply image-specific Neural Architecture Search (NAS) in order to find the optimal architecture within a search space. To that end, we introduced a new NAS approach for DIP settings.
The project aims to develop a GUI for facilitating the experimentation of RL experiments. Users are able to test and implement their RL algorithms via visual drag and drop GUI. I successfully completed the GSoC 2021 program.
University of Tuebingen - Explainable Machine Learning Group
Undergraduate Researcher
Jan 2021 – Mar 2021 Tuebingen (Remote)
Implemented state-of-the-art generators and adapted it to the few-shot learning model proposed in the paper Feature Generating Networks for Zero-Shot Learning and f-VAEGAN-D2: A Feature Generating Framework for Any-Shot Learning to generate better feature representations for unseen images and reported the results.
Bogazici University - Nanonetworking Research Group
Undergraduate Researcher
Nov 2020 – Aug 2021 Istanbul
Proposed the usage of machine learning based receiver design for nano-molecular communication index modulation scenario and conducted experiments using MATLAB. Analysed the efficiacy of CNNs, LSTMs and CNN-LSTMs in solving multivariate time series classification tasks.
University of Cambridge - Affective Intelligence & Robotics Lab
Undergraduate Researcher
Jul 2020 – Oct 2020 Cambridge (Remote)
Implemented continual learning approaches on facial expression recognition task using two well-established datasets: RAF-DB, BP4D, with PyTorch as well as benchmarked state-of-the-art bias mitigation strategies to show the usefulness of continual learning as a bias mitigation approach for FER.
Summer Intern
Jun 2019 – Aug 2019 Istanbul
Conducted data analysis, data visualization and implemented a web-based log file visualization program with Python


What would you do if your beloved pet ran away from home? Wander around the streets? Ask if anyone saw where your pet went? We came up with a solution to this issue (Peter), consisting of three main parts; finding, caring, and adopting. The former aims to make the processes of finding and tracking lost pets easier and in more efficient way. We have developed a platform, where users are able to upload the photos of pets they find on the streets, which will be then used to make a comparison based on the other previously uploaded reported missing pets using image segmentation and deep learning methods. In addition, thanks to the Peter platform, there exists a chance to help the other owners by taking pictures of wandering animals as people walk down the street. By doing so, people could get a discount on the merchandise in the store present on our website. In this way, not only the protection of animals but also the encouragement of helping other people will be reached. From the caring point of view, we aim to connect the pet owners with vets so that they will be taking care of the pets registered in the system in a systematic manner in terms of health and vaccination checks. Additionally, the platform contains a page for adoption, allowing users to easily adopt stray animals. Apart from similarity comparison, we also implemented QR based search for pet owners and we aim to sell collars with QR codes to ease this process. The identity of the pet can be reached by scanning a QR code.